Isaac Lab 2.3 GA Release

[rthaker01]

NVIDIA Isaac Lab 2.3 is now available.

See how the release, built on Isaac Sim 5.1, delivers enhancements across dexterous manipulation, teleoperation, and learning workflows

The update introduces features for dexterous manipulation, including advanced training techniques and new benchmark environments with suction grippers to better scale reinforcement learning.
Enhanced Teleoperation Support for new robots like the Unitree G1 is included, featuring dexterous retargeting that translates human hand motions to the robot, enabling more efficient skill transfer.
The new SkillGen workflow generates adaptive, collision-free manipulation demonstrations by combining human input with GPU-accelerated motion planning.
Developers can use the NVIDIA COMPASS pipeline to train robots for end-to-end navigation, allowing them to move safely through complex, dynamic spaces.
A new workflow synthesizes data for tasks that combine locomotion and manipulation, enabling the creation of large-scale datasets for training humanoid robots on complex sequences.

To learn more about how Isaac Lab extends GPU-native robotics simulation into large-scale multimodal learning to drive the next wave of breakthroughs in robotics research, read the whitepaper and learn more about the latest features in our blog.