ManiSkill v3.0.0b7 is released

[StoneT2000]

ManiSkill v3.0.0b6 is released! Upgrade by running

pip install --upgrade mani_skill

Major Changes:

• Several new tasks added, including some classic locomotion tasks, humanoid tasks, and quadrupeds. New dense rewards for older tasks have been added.
• New assets are used for some robots and floors that look nicer.
• Several RL baselines have been added, RLPD, RFCL, and TDMPC2
• Memory leak fix caused by constant reconfiguration
• Reverted back to pytorch kinematics for now for IK solving in end-effector controllers as it is more stable and has better platform support
• New floating gripper robot, unitree G1 robot, and unitree go 2 robot
• New documentation on RL benchmarking
• Support for using multiple / select GPUs
• Automatic asset download checks when creating environments. Now if an asset is missing, user will be prompted to download relevant assets.

Full Release notes here:

• https://github.com/haosulab/ManiSkill/releases/tag/v3.0.0b7 (bug fix for 2 issues in b6 version)
• https://github.com/haosulab/ManiSkill/releases/tag/v3.0.0b6

Some demos of the new floor assets and single-scene parallel render features

AnymalC-Reach-v1_rt.mp4

opencabinet_low_closeup_sparsely_packed_v2.mp4

g1_pick_place_apple_in_bowl.mp4