My Experiments with
Ground Truth
Learning Physically Interactive Behaviours for Humanoid Robots
Social robots are perceived not just as machines but as interaction partners. Physical touch plays an important role in social interactions, making it an important skill for social robots to learn. The aim of this project is, therefore, to learn human-like physically interactive behaviours for better acceptance of social robots.
MILD: Multimodal Interactive Latent Dynamics for Learning Human-Robot Interaction. Vignesh Prasad, Dorothea Koert, Ruth Stock-Homburg, Jan Peters, Georgia Chalvatzaki. IEEE-RAS International Conference on Humanoid Robots (Humanoids) 2022
Learning Human-like Hand Reaching for Human-Robot Handshaking. Vignesh Prasad, Ruth Stock-Homburg, Jan Peters. IEEE International Conference on Robotics and Automation (ICRA) 2021
Human-Robot Handshaking: A Review. Vignesh Prasad, Ruth Stock-Homburg, Jan Peters. International Journal of Social Robotics (IJSR) 2021
Advances in Human-Robot Handshaking. Vignesh Prasad, Ruth Stock-Homburg, Jan Peters. International Conference on Social Robotics (ICSR) 2020
Evaluation of the Handshake Turing Test for anthropomorphic Robots. Ruth Stock-Homburg, Jan Peters, Katharina Schneider, Vignesh Prasad, Lejla Nukovic. Companion of the ACM/IEEE International Conference on Human Robot Interaction (HRI), 2020
Image Clustering using Gaussian Mixture Variational Autoencoders
We leverage Gaussian Mixture VAEs to learn representations for image clustering in a purely unsupervised manner. We how how image augmentations with weak supervision can help enhance the learning.
Variational Clustering: Leveraging Variational Autoencoders for Image Clustering. Vignesh Prasad*, Dipanjan Das*, Brojeshwar Bhowmick. IEEE International Joint Conference on Neural Networks (IJCNN), 2020
We build on recent advances in unsupervised learning of depth and ego-motion by incorporating Epipolar constraints to make the learning more geometrically sound.
SfMLearner++: Learning Monocular Depth & Ego-Motion using Meaningful Geometric Constraints. Vignesh Prasad, Brojeshwar Bhowmick. IEEE Winter Conference on Applications of Computer Vision (WACV), 2019
Epipolar Geometry based Learning of Multi-view Depth and Ego-Motion from Monocular Sequences. Vignesh Prasad, Dipanjan Das, Brojeshwar Bhowmick. Indian Conference on Computer Vision, Graphics and Image Processing (ICVGIP), 2018, Best Paper Award
Deep Reinforcement Learning for Autonomous Driving
In this project, we explore the use of Curriculum Learning for Deep RL for the task of learning overtaking behaviours in autonomous driving.
Overtaking Maneuvers in Simulated Highway Driving using Deep Reinforcement Learning, Meha Kaushik, Vignesh Prasad, K. Madhava Krishna, Balaraman Ravindran. IEEE Intelligent Vehicles Symposium (IV), 2018
The aim of this project is to use RL and Inverse RL for learning navigational strategies that prevent Monocular SLAM failure from occurring. We propose a map-agnostic method that is able to improve the longevity of Monocular SLAM.
Learning to prevent Monocular SLAM failure using Reinforcement Learning. Vignesh Prasad*, Karmesh Yadav*, Rohitashva Singh Saurabh, Swapnil Daga, Nahas Pareekutty, K. Madhava Krishna, Balaraman Ravindran, Brojeshwar Bhowmick. Indian Conference on Computer Vision, Graphics and Image Processing (ICVGIP), 2018
Data driven strategies for Active Monocular SLAM using Inverse Reinforcement Learning. Vignesh Prasad*, Rishabh Jangir*, K. Madhava Krishna, Balaraman Ravindran. International Conference on Autonomous Agents and Multiagent Systems (AAMAS), 2017
Learning Effective Navigational Strategies for Active Monocular Simultaneous Localization and Mapping. Vignesh Prasad. Master's Thesis, International Institute of Information Technology Hyderabad, 2017.
We aim to detect objects of interests and subsequently reconstruct these objects in teh environment in an efficient manner by running SLAM in parallel. The mapping of the object in SLAM was used to perform a 3D reconstruction of the object.
