Santosha Dwivedy - Academia.edu (original) (raw)

Uploads

Papers by Santosha Dwivedy

Deep Reinforcement Learning has enabled the learning of policies for complex tasks in partially o... more Deep Reinforcement Learning has enabled the learning of policies for complex tasks in partially observable environments, without explicitly learning the underlying model of the tasks. While such model-free methods achieve considerable performance, they often ignore the structure of task. We present a natural representation of to Reinforcement Learning (RL) problems using Recurrent Convolutional Neural Networks (RCNNs), to better exploit this inherent structure. We define 3 such RCNNs, whose forward passes execute an efficient Value Iteration, propagate beliefs of state in partially observable environments, and choose optimal actions respectively. Backpropagating gradients through these RCNNs allows the system to explicitly learn the Transition Model and Reward Function associated with the underlying MDP, serving as an elegant alternative to classical model-based RL. We evaluate the proposed algorithms in simulation, considering a robot planning problem. We demonstrate the capability...

Lecture Notes in Mechanical Engineering

In this work the health monitoring of the tool is carried out by using bimorph piezoelectric patc... more In this work the health monitoring of the tool is carried out by using bimorph piezoelectric patches on the single point cutting tool in turning process. The patches are mounted on the upper and lower surface of the shank of the tool. The vibration analysis of this system is carried out by modeling the tool as an Euler-Bernoulli cantilever beam subjected to transversal base excitation and periodic axial load. The loading is due to forces exerted by work piece on the tool. Extended Hamilton’s principle is used to obtain the governing equation of motion which has been discretized by using generalized Galerkin’s method to obtain the nonlinear temporal equation of motion. Method of multiple scales is used to investigate nonlinear response, generated voltage due to piezoelectric patches of the system. Two resonance conditions have been studied and it is shown that while the simple resonance condition produces the voltage in the order of micro volt, the principal parametric resonance cond...

This paper deals with the dynamics and control of a two degree of freedom robot arm actuated by p... more This paper deals with the dynamics and control of a two degree of freedom robot arm actuated by pneumatic artificial muscles (PAMs). The high power-weight ratio of PAMs justifies their use as actuators in robotics. To achieve trajectory tracking performance, controllers are constructed based on a dynamic model of the robot arm. Due to the high non-linear dynamics of the robotic system, fuzzy control is used for trajectory tracking tasks. The results in simulation are shown to achieve sufficiently good tracking performance.

J. Comput. Inf. Sci. Eng., 2021

This work aims to estimate the lower-limb joint angles in the sagittal plane using Microsoft Kine... more This work aims to estimate the lower-limb joint angles in the sagittal plane using Microsoft Kinect-based experimental setup and apply an efficient machine learning technique for predicting the same based on kinematic, spatiotemporal, and biological parameters. Ten healthy participants from 19 to 50 years (33 ± 11.24 years) were asked to walk in front of the Kinect camera. Based on the skeleton image, the biomechanical hip, knee, and ankle joint angles of the lower-limb were measured using ni-labview. Thereafter, two Bayesian regularization-based backpropagation multilayer perceptron neural network models were designed to predict the joint angles in the stance and swing phase. The joint angles of two individuals, as a testing dataset, were predicted and compared with the experimental results. The test correlation coefficient for predicted joint angles has shown a promising effect of the proposed neural network models. Finally, a qualitative comparison was presented between the joint...

The present paper deals with the design and development of a modular three wheel robot capable of... more The present paper deals with the design and development of a modular three wheel robot capable of autonomously navigating a structured environment by means of tracking a curved line on the floor. Furthermore, it is equipped with a deployable scissor lift with an end effecter plate thus providing applicative modularity to the robot. The main application is envisaged to serve as an automatic guided vehicle in industries such as automotive and construction where cost effectiveness is an essential criterion for automation and mobility is a highly desirable functionality. Mechanical design is carried out using stress analysis with the help of the finite element software package ANSYS and velocity response curves are generated for the highly non-linear motion of the scissor lift. Finally an algorithm is developed for high speed line tracking using LED and phototransistor pairs and results are presented in terms of experimental data as well as simulated responses.

International Journal of Metalcasting

Proceedings of the Advances in Robotics 2019

In this work, a novel mechanized medical injection platform is modeled for automated drug deliver... more In this work, a novel mechanized medical injection platform is modeled for automated drug delivery and phlebotomy. The objective is to design a versatile device that can administer injections at various sites on the human body while being either bed-top or desktop. The machine consists of 6 degrees-of-freedom system including an independent end effector. The design is modeled in the SolidWorks software with required specifications. Thereafter, static structural analysis is carried out for the most critical component. Furthermore, the mathematical modeling is done in MATLAB to estimate the actuator torque for the end effector.

MATEC Web of Conferences, 2018

Journal of Sound and Vibration

Abstract In this work, the nonlinear dynamics of the tool and the thin cylindrical workpiece are ... more Abstract In this work, the nonlinear dynamics of the tool and the thin cylindrical workpiece are studied simultaneously for turning operation. To capture the flexibility of thin workpiece, structural nonlinearity is taken into consideration. Regenerative chatter effect in turning operation is considered in terms of effective chip thickness which is driven by relative motion of the tool and the workpiece during the present cut and the earlier cut. Higher order method of multiple scales (MMS) is used to study the nonlinear responses and stability of tool and workpiece for both internal resonance and primary resonance conditions. The critical values of the cutting parameters like spindle speed, chip width (or depth of cut), cutting force coefficient, workpiece to tool stiffness ratio, and workpiece diameter to thickness ratio etc. are estimated using both time and frequency response curves to have stable chatter free turning operation. The effects of stiffness nonlinearities on the workpiece and the tool responses are also investigated.

Soft Computing

Present studies deal with application of finite element method and intelligent soft computing tec... more Present studies deal with application of finite element method and intelligent soft computing techniques viz. neural network (NN) and adaptive neuro-fuzzy inference system (ANFIS) for the prediction of short duration impulse, ramp and hat forces. Two blunt bodies viz. a hemisphere and a blunt cone with cylindrical aft body have been considered in these investigations. Training of the NN and ANFIS has been carried out using one axial acceleration and two normal accelerations obtained from known input forces and a moment. It is shown here that the NN is unable to recover the unknown forces and moment. However, ANFIS-based strategy is found better for prediction of the same forces and moment. Thus, novelty of these studies exists in the assessment and successful implementation of the soft computing techniques like NN and ANFIS for prediction of the unknown short-duration force and moment time histories. These predictions are also cross-checked for the error, and it has been observed that the ANFIS can be used for short-duration force prediction in high-speed aerospace applications.

2016 23rd International Conference on Pattern Recognition (ICPR), 2016

Procedia Engineering, 2016

Robotica

In this paper, an improved adaptive motion-force control approach is introduced to control the co... more In this paper, an improved adaptive motion-force control approach is introduced to control the cooperative manipulators transporting a shared object under limited communication. The adaptive controller is designed based on the backstepping approach to control the motion of the handled object in the presence of uncertainties and external disturbances. Moreover, the force controller is established to maintain constant internal forces. An event-triggered (ET) mechanism is derived based on the Lyapunov analysis to deal with the bandwidth restrictions and maintain the system stability during the cooperative manipulation. The effectiveness of the proposed control scheme is investigated by comparing it with the existing variations of adaptive backstepping control (i.e., traditional and state augmented schemes). Moreover, the designed triggering mechanism is compared with different triggering conditions presented in the literature. The proposed control approach is further validated in a mor...

Journal of Intelligent & Robotic Systems

International Journal of Pharmaceutics

Journal of Dynamic Systems, Measurement, and Control

This paper uses the Floquet theory for tuning the feedback gains to stabilize the tracking errors... more This paper uses the Floquet theory for tuning the feedback gains to stabilize the tracking errors of a revolute-revolute-revolute-prismatic (RRRP) robot moving in a three-dimensional (3D) workspace. This robot is driven by a proportional-integral-derivative (PID) control law, tracking a time-varying trajectory in joint space, without knowledge of any bounds of the inertia matrix and/or Jacobian of the gravity vector. The Floquet theory is used to obtain the values of feedback gains for which the asymptotic stability of the tracking errors is obtained. The numerical results obtained by Floquet theory are verified by the tracking error plots and phase portraits. The obtained results will be very useful for the control of any industrial robot, required to perform repetitive tasks like assembly of parts and inspection of products, amongst others.

International Journal of Social Robotics

International Journal of Non-Linear Mechanics