Parameterized Quasi-Physical Simulators for Enabling Dexterous Manipulation Transfer
By introducing parameterized quasi-physical simulators and a physics curriculum, we enable a dexterous robot hand to accurately track complex human manipulations involving changing contacts, non-trivial object motions, and intricate tool-using.
Achieving Physically Plausible Quadrupedal Locomotion via Differentiable Simulation
Differentiable simulators enable the use of analytic gradients for optimizing contact-rich robotic tasks, such as quadrupedal locomotion, which can improve sample efficiency over traditional reinforcement learning methods.
SliceIt! - A Dual Simulation Framework for Safely Learning Robotic Food Slicing Tasks
A framework for safely and efficiently learning robot food-slicing tasks in a dual simulation environment, combining a high-fidelity cutting simulator and a robotic simulator, to enable better sim-to-real transfer of learning-based control policies.
Differentiable Simulation Platform for Versatile Robotic Manipulation of Thin-Shell Materials
This work introduces ThinShellLab, a fully differentiable simulation platform that enables flexible learning and evaluation of robotic skills for manipulating diverse thin-shell materials with varying properties.
© 2024 by Linnk AI