Vighnesh Vatsal | IIT Bombay (original) (raw)

Uploads

Papers by Vighnesh Vatsal

Mobile manipulators consist of a ground robot base and a mounted robotic arm, with the two compon... more Mobile manipulators consist of a ground robot base and a mounted robotic arm, with the two components typically controlled as separate subsystems. This is enabled by the fact that most mobile bases with three or four-wheeled designs are inherently stable, though lacking in maneuverability. In contrast, dynamically stable mobile bases offer greater agility and safety in crowded human interaction scenarios, though requiring active balancing. In this work, we consider the balancing problem for a Two-Wheeled Inverted Pendulum Mobile Manipulator (TWIP-MM), designed for retail shelf inspection. Using deep reinforcement learning methods (PPO and SAC), we can generate whole-body control strategies that leverage the motion of the robotic arm for in-place stabilization of the base, through a completely model-free approach. In contrast, tuning a standard PID controller requires a model of the robot, and is considered here as a baseline. Compared to PID control in simulation, the RL-based controllers are found to be more robust against changes in initial conditions, variations in inertial parameters, and disturbances applied to the robot.

IEEE robotics and automation letters, Jul 1, 2021

Supernumerary robotic devices in the form of wearable arms enhance a user's reachable workspa... more Supernumerary robotic devices in the form of wearable arms enhance a user's reachable workspace and provide them with additional capabilities. However, the user may experience considerable force and moment loads on their body due to the robot's motion. In this letter, we present a simulation and trajectory planning framework that aims to minimize the load on a user's muscles while operating a Wearable Robotic Forearm (WRF). Using a high-fidelity model of the human arm, we construct a term for biomechanical costs that is subsequently added to the overall cost function for a motion planner. For evaluation, the planner is initialized with shortest paths linearly interpolated between ten start and goal state pairs in the configuration space, as well as with paths optimized for reaction moments using a local search. We find that the biomechanical planner coupled with locally-optimized initialization reduces mean human muscle fiber forces by up to 23.47% compared to the linearly interpolated trajectories.

Wearable robots have become feasible with recent advances in actuators, electronics, and rapid pr... more Wearable robots have become feasible with recent advances in actuators, electronics, and rapid prototyping. We report on the development of a supernumerary wearable robotic forearm for close-range collaborative activities. In order for this device to be effective in its role, both usability and technical challenges must be addressed. We report on studies identifying desirable usage contexts for such a device, followed by an analysis of the advantage it provides and the loads it exerts on the user. Finally, we discuss ongoing work on adapting path-planning, perception, and behavior models to suit this device»s novel interaction scheme.

arXiv (Cornell University), Nov 14, 2020

We describe a novel wearable robotic arm intended for close-range collaborative activities. Resul... more We describe a novel wearable robotic arm intended for close-range collaborative activities. Results from a user-centered iterative design procedure were applied to the development of a prototype, which was then evaluated in terms of workspace volume and loads on the user. Ongoing and future work involves evaluating the human-robot interaction with an autonomous version of the robot in specific scenarios of collaborative assembly and sorting.

2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

2022 IEEE 18th International Conference on Automation Science and Engineering (CASE)

Creating a vision pipeline for different datasets to solve a computer vision task is a complex an... more Creating a vision pipeline for different datasets to solve a computer vision task is a complex and time consuming process. Currently, these pipelines are developed with the help of domain experts. Moreover, there is no systematic structure to construct a vision pipeline apart from relying on experience, trial and error or using template-based approaches. As the search space for choosing suitable algorithms for achieving a particular vision task is large, human exploration for finding a good solution requires time and effort. To address the following issues, we propose a dynamic and data-driven way to identify an appropriate set of algorithms that would be fit for building the vision pipeline in order to achieve the goal task. We introduce a Transformer Architecture complemented with Deep Reinforcement Learning to recommend algorithms that can be incorporated at different stages of the vision workflow. This system is both robust and adaptive to dynamic changes in the environment. Experimental results further show that our method also generalizes well to recommend algorithms that have not been used while training and hence alleviates the need of retraining the system on a new set of algorithms introduced during test time.

Steps involved in user-centred design of the wearable robotic forearm, going from initial concept... more Steps involved in user-centred design of the wearable robotic forearm, going from initial concepts to an evaluated functional design.

2021 Seventh Indian Control Conference (ICC), 2021

Mobile manipulators consist of a ground robot base and a mounted robotic arm, with the two compon... more Mobile manipulators consist of a ground robot base and a mounted robotic arm, with the two components typically controlled as separate subsystems. This is enabled by the fact that most mobile bases with three or four-wheeled designs are inherently stable, though lacking in maneuverability. In contrast, dynamically stable mobile bases offer greater agility and safety in crowded human interaction scenarios, though requiring active balancing. In this work, we consider the balancing problem for a Two-Wheeled Inverted Pendulum Mobile Manipulator (TWIP-MM), designed for retail shelf inspection. Using deep reinforcement learning methods (PPO and SAC), we can generate whole-body control strategies that leverage the motion of the robotic arm for in-place stabilization of the base, through a completely model-free approach. In contrast, tuning a standard PID controller requires a model of the robot, and is considered here as a baseline. Compared to PID control in simulation, the RL-based controllers are found to be more robust against changes in initial conditions, variations in inertial parameters, and disturbances applied to the robot.

Companion of the 2018 ACM/IEEE International Conference on Human-Robot Interaction, 2018

We describe a novel wearable robotic arm intended for close-range collaborative activities. Resul... more We describe a novel wearable robotic arm intended for close-range collaborative activities. Results from a user-centered iterative design procedure were applied to the development of a prototype, which was then evaluated in terms of workspace volume and loads on the user. Ongoing and future work involves evaluating the human-robot interaction with an autonomous version of the robot in specific scenarios of collaborative assembly and sorting.

ArXiv, 2020

We present an analytical solution for the inverse kinematics (IK) of a robotic arm with one prism... more We present an analytical solution for the inverse kinematics (IK) of a robotic arm with one prismatic joint and four revolute joints. This 5-DoF design is a result of minimizing weight while preserving functionality of the device in a wearable usage context. Generally, the IK problem for a 5-DoF robot does not guarantee solutions due to the system being over-constrained. We obtain an analytical solution by applying geometric projections and limiting the ranges of motion for each DoF. We validate this solution by reconstructing randomly sampled end-effector poses, and find position errors below 2 cm and orientation errors below 4 degrees.

Mobile manipulators consist of a ground robot base and a mounted robotic arm, with the two compon... more Mobile manipulators consist of a ground robot base and a mounted robotic arm, with the two components typically controlled as separate subsystems. This is enabled by the fact that most mobile bases with three or four-wheeled designs are inherently stable, though lacking in maneuverability. In contrast, dynamically stable mobile bases offer greater agility and safety in crowded human interaction scenarios, though requiring active balancing. In this work, we consider the balancing problem for a Two-Wheeled Inverted Pendulum Mobile Manipulator (TWIP-MM), designed for retail shelf inspection. Using deep reinforcement learning methods (PPO and SAC), we can generate whole-body control strategies that leverage the motion of the robotic arm for in-place stabilization of the base, through a completely model-free approach. In contrast, tuning a standard PID controller requires a model of the robot, and is considered here as a baseline. Compared to PID control in simulation, the RL-based controllers are found to be more robust against changes in initial conditions, variations in inertial parameters, and disturbances applied to the robot.

IEEE robotics and automation letters, Jul 1, 2021

Supernumerary robotic devices in the form of wearable arms enhance a user's reachable workspa... more Supernumerary robotic devices in the form of wearable arms enhance a user's reachable workspace and provide them with additional capabilities. However, the user may experience considerable force and moment loads on their body due to the robot's motion. In this letter, we present a simulation and trajectory planning framework that aims to minimize the load on a user's muscles while operating a Wearable Robotic Forearm (WRF). Using a high-fidelity model of the human arm, we construct a term for biomechanical costs that is subsequently added to the overall cost function for a motion planner. For evaluation, the planner is initialized with shortest paths linearly interpolated between ten start and goal state pairs in the configuration space, as well as with paths optimized for reaction moments using a local search. We find that the biomechanical planner coupled with locally-optimized initialization reduces mean human muscle fiber forces by up to 23.47% compared to the linearly interpolated trajectories.

Wearable robots have become feasible with recent advances in actuators, electronics, and rapid pr... more Wearable robots have become feasible with recent advances in actuators, electronics, and rapid prototyping. We report on the development of a supernumerary wearable robotic forearm for close-range collaborative activities. In order for this device to be effective in its role, both usability and technical challenges must be addressed. We report on studies identifying desirable usage contexts for such a device, followed by an analysis of the advantage it provides and the loads it exerts on the user. Finally, we discuss ongoing work on adapting path-planning, perception, and behavior models to suit this device»s novel interaction scheme.

arXiv (Cornell University), Nov 14, 2020

We describe a novel wearable robotic arm intended for close-range collaborative activities. Resul... more We describe a novel wearable robotic arm intended for close-range collaborative activities. Results from a user-centered iterative design procedure were applied to the development of a prototype, which was then evaluated in terms of workspace volume and loads on the user. Ongoing and future work involves evaluating the human-robot interaction with an autonomous version of the robot in specific scenarios of collaborative assembly and sorting.

2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

2022 IEEE 18th International Conference on Automation Science and Engineering (CASE)

Creating a vision pipeline for different datasets to solve a computer vision task is a complex an... more Creating a vision pipeline for different datasets to solve a computer vision task is a complex and time consuming process. Currently, these pipelines are developed with the help of domain experts. Moreover, there is no systematic structure to construct a vision pipeline apart from relying on experience, trial and error or using template-based approaches. As the search space for choosing suitable algorithms for achieving a particular vision task is large, human exploration for finding a good solution requires time and effort. To address the following issues, we propose a dynamic and data-driven way to identify an appropriate set of algorithms that would be fit for building the vision pipeline in order to achieve the goal task. We introduce a Transformer Architecture complemented with Deep Reinforcement Learning to recommend algorithms that can be incorporated at different stages of the vision workflow. This system is both robust and adaptive to dynamic changes in the environment. Experimental results further show that our method also generalizes well to recommend algorithms that have not been used while training and hence alleviates the need of retraining the system on a new set of algorithms introduced during test time.

Steps involved in user-centred design of the wearable robotic forearm, going from initial concept... more Steps involved in user-centred design of the wearable robotic forearm, going from initial concepts to an evaluated functional design.

2021 Seventh Indian Control Conference (ICC), 2021

Mobile manipulators consist of a ground robot base and a mounted robotic arm, with the two compon... more Mobile manipulators consist of a ground robot base and a mounted robotic arm, with the two components typically controlled as separate subsystems. This is enabled by the fact that most mobile bases with three or four-wheeled designs are inherently stable, though lacking in maneuverability. In contrast, dynamically stable mobile bases offer greater agility and safety in crowded human interaction scenarios, though requiring active balancing. In this work, we consider the balancing problem for a Two-Wheeled Inverted Pendulum Mobile Manipulator (TWIP-MM), designed for retail shelf inspection. Using deep reinforcement learning methods (PPO and SAC), we can generate whole-body control strategies that leverage the motion of the robotic arm for in-place stabilization of the base, through a completely model-free approach. In contrast, tuning a standard PID controller requires a model of the robot, and is considered here as a baseline. Compared to PID control in simulation, the RL-based controllers are found to be more robust against changes in initial conditions, variations in inertial parameters, and disturbances applied to the robot.

Companion of the 2018 ACM/IEEE International Conference on Human-Robot Interaction, 2018

We describe a novel wearable robotic arm intended for close-range collaborative activities. Resul... more We describe a novel wearable robotic arm intended for close-range collaborative activities. Results from a user-centered iterative design procedure were applied to the development of a prototype, which was then evaluated in terms of workspace volume and loads on the user. Ongoing and future work involves evaluating the human-robot interaction with an autonomous version of the robot in specific scenarios of collaborative assembly and sorting.

ArXiv, 2020

We present an analytical solution for the inverse kinematics (IK) of a robotic arm with one prism... more We present an analytical solution for the inverse kinematics (IK) of a robotic arm with one prismatic joint and four revolute joints. This 5-DoF design is a result of minimizing weight while preserving functionality of the device in a wearable usage context. Generally, the IK problem for a 5-DoF robot does not guarantee solutions due to the system being over-constrained. We obtain an analytical solution by applying geometric projections and limiting the ranges of motion for each DoF. We validate this solution by reconstructing randomly sampled end-effector poses, and find position errors below 2 cm and orientation errors below 4 degrees.