Jeremy L Wyatt | University of Birmingham (original) (raw)

Robot Task Planning by Jeremy L Wyatt

A long-standing goal of AI is to enable robots to plan in the face of uncertain and incomplete in... more A long-standing goal of AI is to enable robots to plan in the face of uncertain and incomplete information, and to handle task failure intelligently. This paper shows how to achieve this. There are two central ideas. The first idea is to organize the robot's knowledge into three layers: instance knowledge at the bottom, commonsense knowledge above that, and diagnostic knowledge on top. Knowledge in a layer above can be used to modify knowledge in the layer(s) below. The second idea is that the robot should represent not just how its actions change the world, but also what it knows or believes. There are two types of knowledge effects the robot's actions can have: epistemic effects (I believe X because I saw it) and assumptions (I'll assume X to be true). By combining the knowledge layers with the models of knowledge effects, we can simultaneously solve several problems in robotics: (i) task planning and execution under uncertainty; (ii) task planning and execution in open worlds; (iii) explaining task failure; (iv) verifying those explanations. The paper describes how the ideas are implemented in a three-layer architecture on a mobile robot platform. The robot implementation was evaluated in five different experiments on object search, mapping, and room categorization.

Robots must perform tasks efficiently and reliably while acting under uncertainty. One way to ach... more Robots must perform tasks efficiently and reliably while acting under uncertainty. One way to achieve efficiency is to give the robot common-sense knowledge about the structure of the world. Reliable robot behaviour can be achieved by modelling the uncertainty in the world probabilistically. We present a robot system that combines these two approaches and demonstrate the improvements in efficiency and reliability that result. Our first contribution is a probabilistic relational model integrating common-sense knowledge about the world in general, with observations of a particular environment. Our second contribution is a continual planning system which is able to plan in the large problems posed by that model, by automatically switching between decision-theoretic and classical procedures. We evaluate our system on object search tasks in two different real-world indoor environments. By reasoning about the trade-offs between possible courses of action with different informa-tional effects, and exploiting the cues and general structures of those environments, our robot is able to consistently demonstrate efficient and reliable goal-directed behaviour.

PROCEEDINGS OF THE NATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE, Jul 22, 2007

We present integration mechanisms for combining heterogeneous components in a situated informatio... more We present integration mechanisms for combining heterogeneous components in a situated information processing system, illustrated by a cognitive robot able to collaborate with a human and display some understanding of its surroundings. These mechanisms include an architectural schema that encourages parallel and incremental information processing, and a method for binding information from distinct representations that when faced with rapid change in the world can maintain a coherent, though distributed, view of it ...

This paper describes an architecture that combines the complementary strengths of declarative pro... more This paper describes an architecture that combines the complementary strengths of declarative programming and probabilistic graphical models to enable robots to represent, reason with, and learn from, qualitative and quantitative descriptions of uncertainty and knowledge. An action language is used for the low-level (LL) and high-level (HL) system descriptions in the architecture, and the definition of recorded histories in the HL is expanded to allow prioritized defaults. For any given goal, tentative plans created in the HL using default knowledge and commonsense reasoning are implemented in the LL using probabilistic algorithms, with the corresponding observations used to update the HL history. Tight coupling between the two levels enables automatic selection of relevant variables and generation of suitable action policies in the LL for each HL action, and supports reasoning with violation of defaults, noisy observations and unreliable actions in large and complex domains. The architecture is evaluated in simulation and on physical robots transporting objects in indoor domains; the benefit on robots is a reduction in task execution time of 39% compared with a purely probabilistic, but still hierarchical, approach.

There are many different approaches to building a system that can engage in autonomous mental dev... more There are many different approaches to building a system that can engage in autonomous mental development. In this paper we present an approach based on what we term self-understanding, by which we mean the explicit representation of and reasoning about what a system does and doesn't know, and how that knowledge changes under action. We present an architecture and a set of representations used in two robot systems that exhibit a limited degree of autonomous mental development, which we term self-extension. The contributions include: representations of gaps and uncertainty for specific kinds of knowledge, and a goal management and planning system for setting and achieving learning goals.

Flexible, general-purpose robots need to autonomously tailor their sensing and information proces... more Flexible, general-purpose robots need to autonomously tailor their sensing and information processing to the task at hand. We pose this challenge as the task of planning under uncertainty. In our domain, the goal is to plan a sequence of visual operators to apply on regions of interest (ROIs) in images of a scene, so that a human and a robot can jointly manipulate and converse about objects on a tabletop. We pose visual processing management as an instance of probabilistic sequential decision making, and specifically as a Partially Observable Markov Decision Process (POMDP). The POMDP formulation uses models that quantitatively capture the unreliability of the operators and enable a robot to reason precisely about the trade-offs between plan reliability and plan execution time. Since planning in practical-sized POMDPs is intractable, we partially ameliorate this intractability for visual processing by defining a novel hierarchical POMDP based on the cognitive requirements of the corresponding planning task. We compare our hierarchical POMDP planning system (HiPPo) with a non-hierarchical POMDP formulation and the Continual Planning (CP) framework that handles uncertainty in a qualitative manner. We show empirically that HiPPo and CP outperform the naive application of all visual operators on all ROIs. The key result is that the POMDP methods produce more robust plans than CP or the naive visual processing. In summary, visual processing problems

Flexible general purpose robots need to tailor their visual pro- cessing to their task, on the fl... more Flexible general purpose robots need to tailor their visual pro- cessing to their task, on the fly. We propose a new approach to this within a planning framework, where the goal is to plan a sequence of visual operators to apply to the regions of interest (ROIs) in a scene. We pose the visual processing problem as a Partially Observable Markov Decision Process (POMDP). This requires probabilistic models of operator effects to quan- titatively capture the unreliability of the processing actions, and thus reason precisely about trade-offs between plan ex- ecution time and plan reliability. Since planning in practical sized POMDPs is intractable we show how to ameliorate this intractability somewhat for our domain by defining a hier- archical POMDP. We compare the hierarchical POMDP ap- proach with a Continual Planning (CP) approach. On a real robot visual domain, we show empirically that all the plan- ning methods outperform naive application of all visual op- erators. The key result ...

Michael Brenner, Richard Dearden, Charles Gretton, Patrick Eyerich, Thomas Keller, Bernhard Nebel, 2008

One major challenge to the widespread deployment of mobile robots is the ability to autonomously ... more One major challenge to the widespread deployment of mobile robots is the ability to autonomously tailor the sensory processing to the task on hand. In our prior work (Sridharan, Wyatt, and Dearden 2008), we proposed an approach for such general-purpose processing of visual input in an application domain where a robot and a human jointly converse about and manipulate objects on a tabletop by processing the regions of interest (ROIs) in input images. We posed the visual processing management problem as a partially observable ...

Exploration for robotic mapping is typically handled using greedy entropy reduction. Here we show... more Exploration for robotic mapping is typically handled using greedy entropy reduction. Here we show how to apply information lookahead planning to a challenging instance of this problem in which an Autonomous Underwater Vehicle (AUV) maps hydrothermal vents. Given a simulation of vent behaviour we derive an observation function to turn the planning for mapping problem into a POMDP. We test a variety of information state MDP algorithms against greedy, systematic and reactive search strategies. We show that directly ...

Proceedings of the Eleventh Conference Towards Autonomous Robotic Systems (TAROS-10), 2010

Abstract—The problem of where a mobile robot should go to efficiently build a map of its surround... more Abstract—The problem of where a mobile robot should go to efficiently build a map of its surroundings is frequently addressed using entropy reduction techniques. However, in exploration problems where the goal is to find an object or objects of interest, such techniques can be a useful heuristic but are optimising the wrong quantity. An example of such a problem is an autonomous underwater vehicle (AUV) searching the sea floor for hydrothermal vents. The state of the art in these problems is information lookahead in the ...

2010 IEEE/OES Autonomous Underwater Vehicles, 2010

Robot Manipulation by Jeremy L Wyatt

International Journal of Robotics Research, 2017

This paper presents a method for one-shot learning of dexterous grasps, and grasp generation for ... more This paper presents a method for one-shot learning of dexterous grasps, and grasp generation for novel objects. A model of each grasp type is learned from a single kinesthetic demonstration, and several types are taught. These models are used to select and generate grasps for unfamiliar objects. Both the learning and generation stages use an incomplete point cloud from a depth camera – no prior model of object shape is used. The learned model is a product of experts, in which experts are of two types. The first is a contact model and is a density over the pose of a single hand link relative to the local object surface. The second is the hand configuration model and is a density over the whole hand configuration. Grasp generation for an unfamiliar object optimises the product of these two model types, generating thousands of grasp candidates in under 30 seconds. The method is robust to incomplete data at both training and testing stages. When several grasp types are considered the method selects the highest likelihood grasp across all the types. In an experiment, the training set consisted of five different grasps, and the test set of forty-five previously unseen objects. The success rate of the first choice grasp is 84.4% or 77.7% if seven views or a single view of the test object are taken, respectively.

Generalising dexterous grasps to novel objects is an open problem. We show how to learn grasps fo... more Generalising dexterous grasps to novel objects is an open problem. We show how to learn grasps for high DoF hands that generalise to novel objects, given as little as one demonstrated grasp. During grasp learning two types of probability density are learned that model the demonstrated grasp. The first density type (the contact model) models the relationship of an individual finger part to local surface features at its contact point. The second density type (the hand configuration model) models the whole hand configuration during the approach to grasp. When presented with a new object, many candidate grasps are generated, and a kinematically feasible grasp is selected that maximises the product of these densities. We demonstrate 31 successful grasps on novel objects (an 86% success rate), transferred from 16 training grasps. The method enables: transfer of dexterous grasps within object categories; across object categories; to and from objects where there is no complete model of the object available; and using two different dexterous hands.

Current approaches to visual object class detection mainly focus on the recognition of basic leve... more Current approaches to visual object class detection mainly focus on the recognition of basic level categories, such as cars, motorbikes , mugs and bottles. Although these approaches have demonstrated impressive performance in terms of recognition, their restriction to these categories seems inadequate in the context of embodied, cognitive agents. Here, distinguishing objects according to functional aspects based on object affordances is important in order to enable manipulation of and interaction between physical objects and cognitive agent. In this paper, we propose a system for the detection of functional object classes, based on a representation of visually distinct hints on object affordances (affordance cues). It spans the complete range from tutor-driven acquisition of affordance cues, learning of corresponding object models, and detecting novel instances of functional object classes in real images.

The ability to predict how objects behave during manipulation is an important problem. Models inf... more The ability to predict how objects behave during manipulation is an important problem. Models informed by mechanics are powerful, but are hard to tune. An alternative is to learn a model of the object's motion from data, to learn to predict. We study this for push manipulation. The paper starts by formulating a quasi-static prediction problem. We then pose the problem of learning to predict in two different frameworks: i) regression and ii) density estimation. Our architecture is modular: many simple object and context specific predictors are learned. We show empirically that such predictors outperform a rigid body dynamics engine tuned on the same data. We then extend the density estimation approach using a product of experts. This allows transfer of learned motion models to objects of novel shape, and to novel actions. With the right representation and learning method these transferred models can match the prediction performance of a rigid body dynamics engine for novel objects or actions.

2009 International Conference on Advanced Robotics, Jun 22, 2009

A long-standing goal of AI is to enable robots to plan in the face of uncertain and incomplete in... more A long-standing goal of AI is to enable robots to plan in the face of uncertain and incomplete information, and to handle task failure intelligently. This paper shows how to achieve this. There are two central ideas. The first idea is to organize the robot's knowledge into three layers: instance knowledge at the bottom, commonsense knowledge above that, and diagnostic knowledge on top. Knowledge in a layer above can be used to modify knowledge in the layer(s) below. The second idea is that the robot should represent not just how its actions change the world, but also what it knows or believes. There are two types of knowledge effects the robot's actions can have: epistemic effects (I believe X because I saw it) and assumptions (I'll assume X to be true). By combining the knowledge layers with the models of knowledge effects, we can simultaneously solve several problems in robotics: (i) task planning and execution under uncertainty; (ii) task planning and execution in open worlds; (iii) explaining task failure; (iv) verifying those explanations. The paper describes how the ideas are implemented in a three-layer architecture on a mobile robot platform. The robot implementation was evaluated in five different experiments on object search, mapping, and room categorization.

Robots must perform tasks efficiently and reliably while acting under uncertainty. One way to ach... more Robots must perform tasks efficiently and reliably while acting under uncertainty. One way to achieve efficiency is to give the robot common-sense knowledge about the structure of the world. Reliable robot behaviour can be achieved by modelling the uncertainty in the world probabilistically. We present a robot system that combines these two approaches and demonstrate the improvements in efficiency and reliability that result. Our first contribution is a probabilistic relational model integrating common-sense knowledge about the world in general, with observations of a particular environment. Our second contribution is a continual planning system which is able to plan in the large problems posed by that model, by automatically switching between decision-theoretic and classical procedures. We evaluate our system on object search tasks in two different real-world indoor environments. By reasoning about the trade-offs between possible courses of action with different informa-tional effects, and exploiting the cues and general structures of those environments, our robot is able to consistently demonstrate efficient and reliable goal-directed behaviour.

PROCEEDINGS OF THE NATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE, Jul 22, 2007

We present integration mechanisms for combining heterogeneous components in a situated informatio... more We present integration mechanisms for combining heterogeneous components in a situated information processing system, illustrated by a cognitive robot able to collaborate with a human and display some understanding of its surroundings. These mechanisms include an architectural schema that encourages parallel and incremental information processing, and a method for binding information from distinct representations that when faced with rapid change in the world can maintain a coherent, though distributed, view of it ...

This paper describes an architecture that combines the complementary strengths of declarative pro... more This paper describes an architecture that combines the complementary strengths of declarative programming and probabilistic graphical models to enable robots to represent, reason with, and learn from, qualitative and quantitative descriptions of uncertainty and knowledge. An action language is used for the low-level (LL) and high-level (HL) system descriptions in the architecture, and the definition of recorded histories in the HL is expanded to allow prioritized defaults. For any given goal, tentative plans created in the HL using default knowledge and commonsense reasoning are implemented in the LL using probabilistic algorithms, with the corresponding observations used to update the HL history. Tight coupling between the two levels enables automatic selection of relevant variables and generation of suitable action policies in the LL for each HL action, and supports reasoning with violation of defaults, noisy observations and unreliable actions in large and complex domains. The architecture is evaluated in simulation and on physical robots transporting objects in indoor domains; the benefit on robots is a reduction in task execution time of 39% compared with a purely probabilistic, but still hierarchical, approach.

There are many different approaches to building a system that can engage in autonomous mental dev... more There are many different approaches to building a system that can engage in autonomous mental development. In this paper we present an approach based on what we term self-understanding, by which we mean the explicit representation of and reasoning about what a system does and doesn't know, and how that knowledge changes under action. We present an architecture and a set of representations used in two robot systems that exhibit a limited degree of autonomous mental development, which we term self-extension. The contributions include: representations of gaps and uncertainty for specific kinds of knowledge, and a goal management and planning system for setting and achieving learning goals.

Flexible, general-purpose robots need to autonomously tailor their sensing and information proces... more Flexible, general-purpose robots need to autonomously tailor their sensing and information processing to the task at hand. We pose this challenge as the task of planning under uncertainty. In our domain, the goal is to plan a sequence of visual operators to apply on regions of interest (ROIs) in images of a scene, so that a human and a robot can jointly manipulate and converse about objects on a tabletop. We pose visual processing management as an instance of probabilistic sequential decision making, and specifically as a Partially Observable Markov Decision Process (POMDP). The POMDP formulation uses models that quantitatively capture the unreliability of the operators and enable a robot to reason precisely about the trade-offs between plan reliability and plan execution time. Since planning in practical-sized POMDPs is intractable, we partially ameliorate this intractability for visual processing by defining a novel hierarchical POMDP based on the cognitive requirements of the corresponding planning task. We compare our hierarchical POMDP planning system (HiPPo) with a non-hierarchical POMDP formulation and the Continual Planning (CP) framework that handles uncertainty in a qualitative manner. We show empirically that HiPPo and CP outperform the naive application of all visual operators on all ROIs. The key result is that the POMDP methods produce more robust plans than CP or the naive visual processing. In summary, visual processing problems

Flexible general purpose robots need to tailor their visual pro- cessing to their task, on the fl... more Flexible general purpose robots need to tailor their visual pro- cessing to their task, on the fly. We propose a new approach to this within a planning framework, where the goal is to plan a sequence of visual operators to apply to the regions of interest (ROIs) in a scene. We pose the visual processing problem as a Partially Observable Markov Decision Process (POMDP). This requires probabilistic models of operator effects to quan- titatively capture the unreliability of the processing actions, and thus reason precisely about trade-offs between plan ex- ecution time and plan reliability. Since planning in practical sized POMDPs is intractable we show how to ameliorate this intractability somewhat for our domain by defining a hier- archical POMDP. We compare the hierarchical POMDP ap- proach with a Continual Planning (CP) approach. On a real robot visual domain, we show empirically that all the plan- ning methods outperform naive application of all visual op- erators. The key result ...

Michael Brenner, Richard Dearden, Charles Gretton, Patrick Eyerich, Thomas Keller, Bernhard Nebel, 2008

One major challenge to the widespread deployment of mobile robots is the ability to autonomously ... more One major challenge to the widespread deployment of mobile robots is the ability to autonomously tailor the sensory processing to the task on hand. In our prior work (Sridharan, Wyatt, and Dearden 2008), we proposed an approach for such general-purpose processing of visual input in an application domain where a robot and a human jointly converse about and manipulate objects on a tabletop by processing the regions of interest (ROIs) in input images. We posed the visual processing management problem as a partially observable ...

Exploration for robotic mapping is typically handled using greedy entropy reduction. Here we show... more Exploration for robotic mapping is typically handled using greedy entropy reduction. Here we show how to apply information lookahead planning to a challenging instance of this problem in which an Autonomous Underwater Vehicle (AUV) maps hydrothermal vents. Given a simulation of vent behaviour we derive an observation function to turn the planning for mapping problem into a POMDP. We test a variety of information state MDP algorithms against greedy, systematic and reactive search strategies. We show that directly ...

Proceedings of the Eleventh Conference Towards Autonomous Robotic Systems (TAROS-10), 2010

Abstract—The problem of where a mobile robot should go to efficiently build a map of its surround... more Abstract—The problem of where a mobile robot should go to efficiently build a map of its surroundings is frequently addressed using entropy reduction techniques. However, in exploration problems where the goal is to find an object or objects of interest, such techniques can be a useful heuristic but are optimising the wrong quantity. An example of such a problem is an autonomous underwater vehicle (AUV) searching the sea floor for hydrothermal vents. The state of the art in these problems is information lookahead in the ...

2010 IEEE/OES Autonomous Underwater Vehicles, 2010

International Journal of Robotics Research, 2017

This paper presents a method for one-shot learning of dexterous grasps, and grasp generation for ... more This paper presents a method for one-shot learning of dexterous grasps, and grasp generation for novel objects. A model of each grasp type is learned from a single kinesthetic demonstration, and several types are taught. These models are used to select and generate grasps for unfamiliar objects. Both the learning and generation stages use an incomplete point cloud from a depth camera – no prior model of object shape is used. The learned model is a product of experts, in which experts are of two types. The first is a contact model and is a density over the pose of a single hand link relative to the local object surface. The second is the hand configuration model and is a density over the whole hand configuration. Grasp generation for an unfamiliar object optimises the product of these two model types, generating thousands of grasp candidates in under 30 seconds. The method is robust to incomplete data at both training and testing stages. When several grasp types are considered the method selects the highest likelihood grasp across all the types. In an experiment, the training set consisted of five different grasps, and the test set of forty-five previously unseen objects. The success rate of the first choice grasp is 84.4% or 77.7% if seven views or a single view of the test object are taken, respectively.

Generalising dexterous grasps to novel objects is an open problem. We show how to learn grasps fo... more Generalising dexterous grasps to novel objects is an open problem. We show how to learn grasps for high DoF hands that generalise to novel objects, given as little as one demonstrated grasp. During grasp learning two types of probability density are learned that model the demonstrated grasp. The first density type (the contact model) models the relationship of an individual finger part to local surface features at its contact point. The second density type (the hand configuration model) models the whole hand configuration during the approach to grasp. When presented with a new object, many candidate grasps are generated, and a kinematically feasible grasp is selected that maximises the product of these densities. We demonstrate 31 successful grasps on novel objects (an 86% success rate), transferred from 16 training grasps. The method enables: transfer of dexterous grasps within object categories; across object categories; to and from objects where there is no complete model of the object available; and using two different dexterous hands.

Current approaches to visual object class detection mainly focus on the recognition of basic leve... more Current approaches to visual object class detection mainly focus on the recognition of basic level categories, such as cars, motorbikes , mugs and bottles. Although these approaches have demonstrated impressive performance in terms of recognition, their restriction to these categories seems inadequate in the context of embodied, cognitive agents. Here, distinguishing objects according to functional aspects based on object affordances is important in order to enable manipulation of and interaction between physical objects and cognitive agent. In this paper, we propose a system for the detection of functional object classes, based on a representation of visually distinct hints on object affordances (affordance cues). It spans the complete range from tutor-driven acquisition of affordance cues, learning of corresponding object models, and detecting novel instances of functional object classes in real images.

The ability to predict how objects behave during manipulation is an important problem. Models inf... more The ability to predict how objects behave during manipulation is an important problem. Models informed by mechanics are powerful, but are hard to tune. An alternative is to learn a model of the object's motion from data, to learn to predict. We study this for push manipulation. The paper starts by formulating a quasi-static prediction problem. We then pose the problem of learning to predict in two different frameworks: i) regression and ii) density estimation. Our architecture is modular: many simple object and context specific predictors are learned. We show empirically that such predictors outperform a rigid body dynamics engine tuned on the same data. We then extend the density estimation approach using a product of experts. This allows transfer of learned motion models to objects of novel shape, and to novel actions. With the right representation and learning method these transferred models can match the prediction performance of a rigid body dynamics engine for novel objects or actions.

2009 International Conference on Advanced Robotics, Jun 22, 2009

How should a robot direct active vision so as to ensure reliable grasping? We answer this questio... more How should a robot direct active vision so as to ensure reliable grasping? We answer this question for the case of dexterous grasping of unfamiliar objects. When an object is unfamiliar, much of its shape is by definition unknown. An initial view will recover only some surfaces, leaving most of the object's surface unmodelled, and also leaving shadow regions which may or may not contain obstacles. These two features make it difficult both to select reliable grasps, and to plan safe reach-to-grasp trajectories. Grasps typically fail in one of two ways, either unmodelled objects in the scene cause collisions, or object reconstruction is insufficient to ensure that the grasp points provide a stable force closure. These problems can be solved more easily if active sensing is guided by the anticipated actions. Our approach has three stages. First, we take a single view and generate candidate grasps from the resulting partial object reconstruction. Second, we drive active vision to maximise surface reconstruction quality around the planned contact points. During this phase the anticipated grasp is continually refined. Third, we direct gaze to unmodelled regions that will affect the planned reach to grasp trajectory, so as to confirm that this trajectory is safe. We show, on a dexterous manipulator with camera on wrist, that our approach (85.7% success rate) outperforms a randomised algorithm (48% success rate).

ACM Transactions on Applied Perception, Oct 1, 2013

2012 IEEE International Conference on Robotics and Automation, 2012

2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2012

Abstract—This work addresses the problem of planning the reach-to-grasp trajectory for a robotic ... more Abstract—This work addresses the problem of planning the reach-to-grasp trajectory for a robotic arm and hand, when there is uncertainty in the pose of the object being grasped. If the object is not in its expected location, then the robot may still gain additional information about the object pose by making tactile or haptic observations if a finger or other part of the hand collides with part of the object during the reach-to-grasp operation. Therefore, it is desirable to plan the reach-to-grasp trajectory in such a way that it takes into account and ...

2013 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2013

ABSTRACT Dexterous grasping of objects with uncertain pose is a hard unsolved problem in robotics... more ABSTRACT Dexterous grasping of objects with uncertain pose is a hard unsolved problem in robotics. This paper solves this problem using information gain re-planning. First we show how tactile information, acquired during a failed attempt to grasp an object can be used to refine the estimate of that object’s pose. Second, we show how this information can be used to replan new reach to grasp trajectories for successive grasp attempts. Finally we show how reach-to-grasp trajectories can be modified, so that they maximise the expected tactile information gain, while simultaneously delivering the hand to the grasp configuration that is most likely to succeed. Our main novel outcome is thus to enable tactile information gain planning for Dexterous, high degree of freedom (DoFs) manipulators. We achieve this using a combination of information gain planning, hierarchical probabilistic roadmap planning, and belief updating from tactile sensors for objects with non-Gaussian pose uncertainty in 6 dimensions. The method is demonstrated in trials with simulated robots. Sequential replanning is shown to achieve a greater success rate than single grasp attempts, and trajectories that maximise information gain require fewer re-planning iterations than conventional planning methods before a grasp is achieved.

Abstract. Artificial Intelligence (AI) and Animal Cognition (AC) share a common goal: to study le... more Abstract. Artificial Intelligence (AI) and Animal Cognition (AC) share a common goal: to study learning and causal understanding. However, the perspectives are completely different: while AC studies intelligent systems present in nature, AI tries to to build them almost from scratch. It is proposed here that both visions are complementary and should interact more to better achieve their ends. Nonetheless, before efficient collaboration can take place, a greater mutual understanding of each field is required, beginning with clarifications of their ...

Studies in Applied Philosophy, Epistemology and Rational Ethics, 2013

Research and Development in Intelligent Systems XXVIII, 2011

Recently there has been a good deal of interest in using techniques developed for learning from r... more Recently there has been a good deal of interest in using techniques developed for learning from reinforcement to guide learning in robots. Motivated by the desire to find better robot learning methods, this thesis prsents a number of novel extensions to existing techniques for controlling exploration and inference in reinforcement learning. First I distinguish between the well known exploration-exploitation trade-off and what I term exploration for future exploitation. it is argued that there are many tasks where it is more appropriate to ...

The Journal of Machine Learning …, 2005

Ensembles are a widely used and effective technique in machine learning—their success is com-monl... more Ensembles are a widely used and effective technique in machine learning—their success is com-monly attributed to the degree of disagreement, or 'diversity', within the ensemble. For ensembles where the individual estimators output crisp class labels, this 'diversity' is not well ...

Lecture Notes in Computer Science, 2005

Lecture Notes in Computer Science, 2003

Abstract: Neural network ensembles are well accepted as a route to combining a group of weaker le... more Abstract: Neural network ensembles are well accepted as a route to combining a group of weaker learning systems in order to make a composite, stronger one. It has been shown that low correlation of errors (" diverse members") will give rise to better ensemble ...

Proceedings of the 25th Workshop of the UK Planning and Scheduling Special Interest Group (PlanSIG 2006), Dec 1, 2006

Model transfer refers to the process of transferring information from a model that was previously... more Model transfer refers to the process of transferring information from a model that was previously identified for one task (source) to a new task (target). For decision tasks in unknown Markov environments, we profit through model transfer by using information from related tasks, eg transition knowledge and solution (policy) knowledge, to quickly determine an appropriate model of the new task environment. A difficulty with such transfer is typically the non-linear and indirect relationship between the available source knowledge and the ...

Benelearn 2005 Annual Machine Learning Conference of Belgium and the Netherlands, 2008

An important problem in reinforcement learning is determining how to act while learning sometimes... more An important problem in reinforcement learning is determining how to act while learning sometimes referred to as the exploration-exploitation dilemma or the problem of optimal learning. The problem is intractable, usually solved through approximation such as by being optimistic in the face of uncertainty. In environments with inherent determinism, arising for example from known process templates, acting conforms to certain acceptable conventions that limit exploration. We present an algorithm for the learning problem in ...

Power priors allow us to introduce into a Bayesian algorithm a relative precision parameter that ... more Power priors allow us to introduce into a Bayesian algorithm a relative precision parameter that controls the influence of external evidence on a new task. Such evidence, often available as historical data, can be quite useful when learning a new task from reinforcement. In this paper, we study the use of power priors in Bayesian reinforcement learning. We start by describing the basics of power prior distributions. We then develop power priors for unknown Markov decision processes incorporating historical data. Finally, ...

Planning, Learning and Monitoring with Uncertainty and Dynamic Worlds, Aug 1, 2006

Abstract. We describe techniques for Bayesian inference in Markov Decision Processes (MDP) with u... more Abstract. We describe techniques for Bayesian inference in Markov Decision Processes (MDP) with unknown transition probabilities in which specification of smoothness are provided with the prior distribution. In particular, we consider smoothness conditions expressed as an umbrella type order relations on the transition probabilities, thus restricting the space of conforming process models. We study a transformation approach that obtains samples of process models for a constrained posterior density by mapping draws from an ...

Abstract This paper describes a new reinforcement learning (RL) model, the incremental topologica... more Abstract This paper describes a new reinforcement learning (RL) model, the incremental topological reinforcement learning agent (ITRLA), designed to guide agent navigation in non-structured environments, considering two common situations:(i) insertion of noise during state estimation and (ii) changes in environment structure. Tasks in non-structured environments are hard to be learned by traditional RL algorithms due to the stochastic state transitions. Such tasks are often modeled as partially observable Markov decision ...

The starting point of this position paper is the observation that robot learning of tasks, when d... more The starting point of this position paper is the observation that robot learning of tasks, when done autonomously, can be conveniently divided into three learning problems. In the first we must derive a controller for a task given a process model. In the second we must derive such a process model, perhaps in the face of hidden state. In the third we search for perceptual processing functions that find natural regularities in the robot's sensory sequence, and which have utility in so far as they ease the construction of process models ...

Philosophical Studies, 2017

Slurring is a kind of hate speech that has various effects. Notable among these is variable offen... more Slurring is a kind of hate speech that has various effects. Notable among these is variable offence. Slurs vary in offence across words, uses, and the reactions of audience members. Patterns of offence aren't adequately explained by current theories. We propose an explanation based on the unjust power imbalance that a slur seeks to achieve. Our starting observation is that in discourse participants take on discourse roles. These are typically inherited from social roles, but only exist during a discourse. A slurring act is a speech-act that alters the discourse roles of the target and speaker. By assigning discourse roles the speaker unjustly changes the power balance in the dialogue. This has a variety of effects on the target and audience. We show how these notions explain all three types of offence variation. We also briefly sketch how a role and power theory can help explain silencing and appropriation. Explanatory power lies in the fact that offence is correlated with the perceived unjustness of the power imbalance created.

2014 IEEE International Conference on Robotics and Automation (ICRA), 2014

2011 IEEE International Conference on Robotics and Automation, 2011

2010 IEEE International Conference on Robotics and Automation, 2010

2011 IEEE International Conference on Robotics and Automation, 2011

Perception

Page 1. Gaze Allocation During Visually Guided Manipulation Technical Report Jose Nunez-Varela¹, ... more Page 1. Gaze Allocation During Visually Guided Manipulation Technical Report Jose Nunez-Varela¹, Priya A. Mani², B. Ravindran², Jeremy L.Wyatt¹ ¹School of Computer Science University of Birmingham ²Department of Computer Science and Engineering IIT Madras April ...

Procedings of the British Machine Vision Conference 2011, 2011

Lecture Notes in Computer Science, 2014

Neural Networks and Computational Intelligence, 2003

This paper presents experimental results on image analysis for a particular form of Genetic Progr... more This paper presents experimental results on image analysis for a particular form of Genetic Programming called Cartesian Ge- netic Programming (CGP) in which programs use the structure of a graph represented as a linear sequence of integers. The efficency of this approach is investigated for the problem of Object Localization in a given image. This task is usually car- ried

One major challenge to the widespread deployment of mo- bile robots is the ability to autonomousl... more One major challenge to the widespread deployment of mo- bile robots is the ability to autonomously tailor the sensory processing to the task on hand. In our prior work (Sridharan, Wyatt, and Dearden 2008), we proposed an approach for such general-purpose processing of visual input in an application domain where a robot and a human jointly converse about and manipulate objects on a tabletop by processing the regions of interest (ROIs) in input images. We posed the visual process- ing management problem as a partially observable Markov decision problem (POMDP), and introduced a hierarchical decomposition to make it tractable to plan with POMDPs. In this paper we analyze and eliminate some of the limitations of the existing approach. First, in addition to tackling visual actions that analyze the state of the world represented by the image, we show how to incorporate actions that can change the state. Secondly, we show how policy caching can be used to speed the planning performance and...

Human studies have shown that gaze shifts are mostly driven by the current task demands. In manip... more Human studies have shown that gaze shifts are mostly driven by the current task demands. In manipulation tasks, gaze leads action to the next manipulation target. One explanation is that fixations gather information about task relevant properties, where task relevance is signalled by reward. This work presents new computational models of gaze shifting, where the agent imagines ahead in time the informational effects of possible gaze fixations. Building on our previous work, the contributions of this paper are: a) the presentation of two new gaze control models; b) comparison of their performance to our previous model; c) results showing the fit of all these models to previously published human data; and d) integration of a visual search process. The first new model selects the gaze that most reduces positional uncertainty of landmarks (Unc), and the second maximises expected rewards by reducing positional uncertainty (RU). Our previous approach maximises the expected gain in cumulative reward by reducing positional uncertainty (RUG). In experiment b) the models are tested on a simulated humanoid robot performing a manipulation task, and each model's performance is characterised by varying three environmental variables. This experiment provides evidence that the RUG model has the best overall performance. In experiment c) we compare the hand-eye coordination timings of the models in a robot simulation to those obtained from human data. This provides evidence that only the models that incorporate both uncertainty and reward (RU and RUG) match human data.

Lecture Notes in Computer Science, 2014

This paper describes an architecture that combines the complemen- tary strengths of probabilistic... more This paper describes an architecture that combines the complemen- tary strengths of probabilistic graphical models and declarative programming to enable robots to represent and reason with qualitative and quantitative descrip- tions of uncertainty and domain knowledge. An action language is used for the architecture’s low-level (LL) and high-level (HL) system descriptions, and the HL definition of recorded history is expanded to allow prioritized defaults. For any given objective, tentative plans created in the HL using commonsense reasoning are implemented in the LL using probabilistic algorithms, and the correspond- ing observations are added to the HL history. Tight coupling between the levels helps automate the selection of relevant variables and the generation of policies in the LL for each HL action, and supports reasoning with violation of defaults, noisy observations and unreliable actions in complex domains. The architecture is evaluated in simulation and on robots moving objects in indoor domains.

ABSTRACTThis paper describes an architecture that combines the complementary strengths of probabi... more ABSTRACTThis paper describes an architecture that combines the complementary strengths of probabilistic graphical models and declarative programming to enable robots to represent and reason with logic-based and probabilistic descriptions of uncertainty and domain knowledge. An action language is extended to support non-boolean fluents and non-deterministic causal laws. This action language is used to describe tightly-coupled transition diagrams at two levels of granularity, refining a coarse-resolution transition diagram of the domain to obtain a fine-resolution transition diagram. The coarse-resolution system description, and a history that includes (prioritized) defaults, are translated into an Answer Set Prolog (ASP) program. For any given goal, inference in the ASP program provides a plan of abstract actions. To implement each such abstract action probabilistically, the part of the fine-resolution transition diagram relevant to this action is identified, and a probabilistic representation of the uncertainty in sensing and actuation is included and used to construct a partially observable Markov decision process (POMDP). The policy obtained by solving the POMDP is invoked repeatedly to implement the abstract action as a sequence of concrete actions, with the corresponding observations being recorded in the coarse-resolution history and used for subsequent reasoning. The architecture is evaluated in simulation and on a mobile robot moving objects in an indoor domain, to show that it supports reasoning with violation of defaults, noisy observations and unreliable actions, in complex domains.

Proceedings of the 3rd international conference on Human robot interaction - HRI '08, 2008

Aaai Workshop on Goal Directed Autonomy, 2010

Cognitive Systems Monographs, 2010

The study of architectures to support intelligent behaviour is certainly the broadest, and arguab... more The study of architectures to support intelligent behaviour is certainly the broadest, and arguably one of the most ill-defined enterprises in AI and Cognitive Science. The basic scientific question we seek to answer is:“What are the trade-offs between the different ways that intelligent systems might be structured?” These trade-offs depend in large part on what kinds of tasks and environment a system operates under (niche space), and also what aspects of the design space we deem to be architectural. In CoSy we have tried to answer ...

Evaluating Architectures for Intelligence: Papers from the 2007 AAAI Workshop, 2007

In this paper we propose an empirical method for the comparison of architectures designed to prod... more In this paper we propose an empirical method for the comparison of architectures designed to produce similar behaviour from an intelligent system. The approach is based on the exploration of design space using similar designs that all satisfy the same requirements in niche space. An example of a possible application of this method is given using a robotic system that has been implemented using a software toolkit that has been designed to support architectural experimentation.

Knowledge-Based Systems, 2009

Thanks to the efforts of the robotics and autonomous systems community, the myriad applications a... more Thanks to the efforts of the robotics and autonomous systems community, the myriad applications and capacities of robots are ever increasing. There is increasing demand from end users for autonomous service robots that can
operate in real environments for extended periods. In the Spatiotemporal Representations and Activities for Cognitive Control in Long-Term Scenarios (STRANDS) project (http://strands- project.eu), we are tackling this demand head-on by integrating state-of-the-art artificial intelligence and robotics research
into mobile service robots and deploying
these systems for long-term installations in security and care environments. Our robots
have been operational for a combined duration of 104 days over four deployments, autonomously performing end-user-defined tasks and traversing 116 km in the process. In this article, we describe the approach we used to enable long-term autonomous operation in everyday environments and how our robots are able to use their long run times to improve their own performance.

We present integration mechanisms for combining heterogeneous components in a situated informatio... more We present integration mechanisms for combining heterogeneous components in a situated information processing system, illustrated by a cognitive robot able to collaborate with a human and display some understanding of its surroundings. These mechanisms include an architectural schema that encourages parallel and incre-mental information processing, and a method for binding information from distinct representations that when faced with rapid change in the world can maintain a coherent , though distributed, view of it. Provisional results are demonstrated in a robot combining vision, manipulation , language, planning and reasoning capabilities interacting with a human and manipulable objects.

International Joint Conference on Artificial Intelligence, 2009

Exploration for robotic mapping is typically handled using greedy entropy reduction. Here we show... more Exploration for robotic mapping is typically handled using greedy entropy reduction. Here we show how to apply information lookahead planning to a challenging instance of this problem in which an Autonomous Underwater Vehicle (AUV) maps hydrothermal vents. Given a simulation of vent behaviour we derive an observation function to turn the planning for mapping problem into a POMDP. We test

Advanced Engineering Informatics, 2010

Proc. IJCAI, Jan 6, 2007

In human-robot interaction (HRI) it is essential that the robot interprets and reacts to a human&... more In human-robot interaction (HRI) it is essential that the robot interprets and reacts to a human's utterances in a manner that reflects their intended meaning. In this paper we present a collection of novel techniques that allow a robot to interpret and execute spoken commands describing manipulation goals involving qualitative spatial constraints (eg “put the red ball near the blue cube”). The resulting implemented system integrates computer vision, potential field models of spatial relationships, and action planning to mediate ...

Cognitive Robotics: Papers from the 2006 AAAI Workshop: Technical Report WS-06-03, http://www. aaai. org/Library/Workshops/ws06-03. php, 2006

This paper discusses some of the long term objectives of cognitive robotics and some of the requi... more This paper discusses some of the long term objectives of cognitive robotics and some of the requirements for meeting those objectives that are still a very long way off. These include requirements for visual perception, for architectures, for kinds of learning, and for innate competences needed to drive learning and development in a variety of different environments. The work arises mainly out of research on requirements for forms of representation and architectures within the PlayMate scenario, which is a scenario ...

2006 6th IEEE-RAS International Conference on Humanoid Robots, 2006

Lecture Notes in Computer Science, 2008

Lecture Notes in Computer Science, 2003

Cognitive Systems Monographs, 2010

Research in CoSy was scenario driven. Two scenarios were created, the Play-Mate and the Explorer.... more Research in CoSy was scenario driven. Two scenarios were created, the Play-Mate and the Explorer. One of the integration goals of the project was to build integrated systems that addressed the tasks in these two scenarios. This chapter concerns the integrated system for the PlayMate scenario.

Proceedings Ieee International Conference on Robotics and Automation, May 9, 2011

Journal of Machine Learning Research, 2004

ABSTRACT

Ensemble approaches to classification and regression have attracted a great deal of interest in r... more Ensemble approaches to classification and regression have attracted a great deal of interest in recent years. These methods can be shown both theoretically and empirically to outperform single predictors on a wide range of tasks. One of the elements required for accurate prediction when using an ensemble is recognised to be error ''diversity''. However, the exact meaning of this concept is not clear from the literature, particularly for classification tasks. In this paper we first review the varied attempts to provide a formal explanation of error diversity, including several heuristic and qualitative explanations in the literature. For completeness of discussion we include not only the classification literature but also some excerpts of the rather more mature regression literature, which we believe can still provide some insights. We proceed to survey the various techniques used for creating diverse ensembles, and categorise them, forming a preliminary taxonomy of diversity creation methods. As part of this taxonomy we introduce the idea of implicit and explicit diversity creation methods, and three dimensions along which these may be applied. Finally we propose some new directions that may prove fruitful in understanding classification error diversity.

International Conference on Automated Planning and Scheduling (ICAPS), Sep 14, 2008

Flexible general purpose robots need to tailor their visual processing to their task, on the fly.... more Flexible general purpose robots need to tailor their visual processing to their task, on the fly. We propose a new approach to this within a planning framework, where the goal is to plan a sequence of visual operators to apply to the regions of interest (ROIs) in a scene. We pose the visual processing problem as a Partially Observable Markov Decision Process (POMDP). This requires probabilistic models of operator effects to quantitatively capture the unreliability of the processing actions, and thus reason precisely about trade-offs between ...

2008 AAAI Fall Symposium on Biologically Inspired Cognitive Architectures, 2008

In this paper we describe insights for theories of natural intelligence that arise from recent ad... more In this paper we describe insights for theories of natural intelligence that arise from recent advances in architectures for robot intelligence. In particular we advocate a sketch theory for the study of both natural and artificial intelligence that consists of a set of constraints on architectures. The sketch includes the use of multiple shared workspaces, parallel asynchronous refinement of shared representations, statistical integration of evidence within and across modalities, massively parallel prediction and content ...

This paper discusses our views on the future of the eld of cognitive architectures, and how the s... more This paper discusses our views on the future of the eld of cognitive architectures, and how the scien- tic questions that dene it should be addressed. We also report on a set of requirements, and a related architecture design, that we are currently investigating as part of the CoSy project. 1 What Are Architectures? The rst problem we face as researchers in the eld of cognitive architectures is dening exactly what we are studying. This is important because the term ìarchitectureî is so widely used in modern techno- logical elds. An agent's cognitive architecture de- nes the information-processing components within the ìmindî of the agent, and how these components are structured in relation to each other. Also, there is a close link between architectures and the mech- anisms and representations used within them (where representations can be of many kinds with many func- tions). Langley and Laird (2002) describe a cognitive architecture as including ìthose aspects of a cognitive ...

Cognitive Systems Monographs, 2010

ABSTRACT

SCHOOL OF COMPUTER SCIENCE RESEARCH REPORTS-UNIVERSITY OF BIRMINGHAM CSR, Nov 24, 2006

DRAFT: Please do not quote without permission; comments welcome. ... 2 Architectures and the Scie... more DRAFT: Please do not quote without permission; comments welcome. ... 2 Architectures and the Science of Cognitive Systems ... 7.1 The Motive Generator . . . . . . . . . . . . . . . . . . . . . . . . . . 12 7.2 The Global Goal Manager and Subarchitecture Task Managers . . . . 12 7.3 TheGeneralMemory . . . . . . . . . . . . . . . . . . . . . . . . . . 14 ... 11 Scenario-Specific Instantiations of the Architecture Schema ... 11.1 APlayMateInstantiation . . . . . . . . . . . . . . . . . . . . . . . . 22 11.2 AnExplorerInstantiation . . . . . . . . . . . . . . . . . . . . . . . . 23 11.3 A Plan Generation Subarchitecture in Detail . . . . . . . . . . . . ...

Third International Conference on Natural Computation (ICNC 2007), 2007

Abstract A computational method for implementation of Evolution Strategies (ES) in Grid computing... more Abstract A computational method for implementation of Evolution Strategies (ES) in Grid computing environments is discussed. In this paper, list scheduling with Round-robin order Replication (RR) is adopted to reduce waiting times due to synchronization in Medium-grained ES. Our results show that the replication in RR can reduce the synchronous waiting time in comparison with Work Queue (WQ) methods.

Advances in Cognitive Neurodynamics ICCN 2007, 2008

Task scheduling algorithms for evolving artificial neural networks (EANNs) in grid computing envi... more Task scheduling algorithms for evolving artificial neural networks (EANNs) in grid computing environments is discussed. In this paper, list scheduling with round-robin order replication (RR) is adopted to reduce waiting times due to synchronization. However, RR is suitable for coarse-grained tasks. For EANNs as medium-grained tasks, we propose a new technique to reduce the communication overhead, called the remote work queue (RWQ) method. We then define round-robin replication remote work queue (R3Q) as RWQ with RR. Our ...