Project 58: Design, Optimization, and Deployment of In-Hand Object Rotation Control
Contact Information
Prof. Wang Hesheng
Email: wanghesheng@sjtu.edu.cn
Project Description and Objectives
The task of in-hand object rotation enhances robotic manipulation capabilities, enabling the execution of more sophisticated operations by integrating tactile and visual sensory information. However, controlling multi-degree-of-freedom systems, processing complex sensory data, and manipulating objects with uncertain motion dynamics present substantial challenges. The rapid advancements in reinforcement learning provide a promising avenue for addressing these challenges, offering robust adaptability across diverse tasks.
This study aims to leverage reinforcement learning to model robotic states within a simulated environment, design task-specific reward functions to optimize training processes, and enable biomimetic robotic hands to perform complex in-hand manipulation tasks. To accomplish these objectives, participants are expected to acquire a comprehensive understanding of reinforcement learning principles, develop training environments tailored to controller design, and validate the proposed approach through experimental deployment on real-world robotic systems.
Eligibility Requirements
Python, ROS
Main Tasks
Design an In-Hand Object Rotation Controller, Relying on Tactile and Visual Information for Training Dexterous Hand Manipulation Tasks
Conduct real-world experiments to achieve sim-to-real transfer, enabling a dexterous robotic hand to perform object rotation tasks in practical settings.
Website