Visual Next-Frame Prediction using Multisensory Perception for Embodied Agents

Robot kinematics, dynamics, control and learning

Natural Locomotion, Jumping and Recovery of Quadruped Robot A1 with AMP

A deep reinforcement learning agent learns to navigate and collect rewards in a large world using lidar and camera.

A product of my Summer 2023 Internship at NYU Courant

A Framework for Sensorimotor Cross-Perception and Cross-Behavior Knowledge Transfer for Object Categorization

Papers, codes, datasets, applications, tutorials.

[IROS'23] Value-Informed Skill Chaining for Policy Learning of Long-Horizon Tasks with Surgical Robot

Labs for the Robot Learning class, focusing on robotics and Reinforcement Learning. Each lab focuses on a different topic, had mandatory tasks and eventually extensions. All the results have been discussed in the reports.

In developmental robotics, robot learning algorithms generate their own sequences of learning experiences, also known as a curriculum, to cumulatively acquire new skills through self-guided exploration and social interaction with humans. These robots use guidance mechanisms such as active learning, maturation, motor synergies and imitation. Asso…

Implementation of Soft QLearning Algorithm

A Framework for Multisensory Foresight for Embodied Agents

Experiment code for "Koopman Constrained Policy Optimization: a Koopman operator theoretic method for differentiable optimal control in robotics" as presented at ICML 2023

Topics include function approximation, learning dynamics, using learned dynamics in control and planning, handling uncertainty in learned models, learning from demonstration, and model-based and model-free reinforcement learning.

This repository hosts the physical robot code for ToolFlowNet. Published at CoRL '22.

Webots Image Dataset Collection For Computer Vision And Deep Learning

Lunar Lander game from OpenAI Gym using behavioral cloning, DAgger methods, and POMDP(Partially-Observable Markov Decision Processes)

Using DAgger with our MPC treated as the expert, we are able to effectively distill knowledge into relatively simple networks while still being able to retain a large fraction of the performance. (Please see paper for full description).

Domain Randomization for Robust, Affordable and Effective Closed-loop Control of Soft Robots