Welcome to try our ManiSkill2 with Tianshou

[Jiayuan-Gu]

Thanks for the great RL framework Tianshou! We are going to provide an example of how to use Tianshou to train a vision-based RL policy on our physically realistic environments in ManiSkill2.

We are announcing ManiSkill2. ManiSkill2 is a large-scale robotic manipulation benchmark, focusing on learning generalizable robot agents and manipulation skills. It features 2000+ diverse objects, 20 task categories, and a large-scale demonstration set in SAPIEN, a fully-physical, realistic simulator. The benchmark can be used to study 2D & 3D vision-based imitation learning, reinforcement learning, and motion planning, etc. We invite you to participate in the associated ManiSkill 2022 challenge where we will be awarding prizes to the teams who achieve the highest success rates in our environments.