Chenrui Tie

I am a last-year undergraduate student in Center on Frontiers of Computing Studies (CFCS) at Peking University, collaborated with Professor Hao Dong and Professor He Wang, now I'm working with Professor Lin Shao in NUS .

My research focuses on constructing models by incorporating varying degrees of prior physics knowledge based on the specific characteristics of different robot manipulation tasks.

Email: crtie [at] pku.edu.cn

Email  /  Github

We proposed a new task set, including a class of common "soft insertion" tasks in real life, where both the object to be inserted and the container can be deformable. We utilized a differentiable simulator design framework to enable robots to acquire the skills rapidly.

We developed Jade, a differentiable physics engine for articulated rigid bodies, using the continuous collision detection to detect the time of impact for collision checking and adopt the backtrack strategy to prevent intersection between bodies with complex geometry shapes. We also derived the gradient calculation to ensure the whole simulation process differentiable under the backtrack mechanism.

We tackle multi-part geometrically assembly task, leveraging SE(3) equivariance and invariance, which fits the physics of the task and narrows the solution space. Thus our method performs better than previous methods in two datasets, and is intuitive to future works.

• Improvement and application of differential simulator
• Sim2real adaptation
• Supervisor: Prof. Lin Shao

• SE(3) equivariant geometric part assembly
• Supervisor: Prof. Hao Dong
• Collaborated with Ruihai Wu

• Model based reinforcement learning for articulated object manipulation on real robot
• Supervisor: Prof. He Wang
• Familiarity with common reinforcement learning algorithms and the control of the Franka robot arm

• Physics intuitive learning
• Supervisor: Prof. He Wang

• Advisor: Prof. Hao Dong, Prof. He Wang
• Lab: Hyperplane Lab and Epic Lab, Center on Frontiers of Computing Studies