isaac-sim/isaaclab is a GPU-accelerated simulation framework for robot learning. It sets up robot environments and learning experiments on top of physics and rendering backends, and the repository description calls it an integrated framework that supports multiple backends. The README says it unifies and simplifies robot research workflows such as reinforcement learning, imitation learning, and post-training. The most recent release is v3.0.0.
Reinforcement learning is a learning method that acquires behavior by following reward signals, while imitation learning learns by following demonstrated motions. Bundling post-training into the same flow is the goal of this framework. It provides the robot learning environments and the physics and sensor simulation needed, together with GPU acceleration, and supports Newton, OVRTX, OVPhysX, and NVIDIA Isaac Sim as physics and rendering backends. It also presents options for running locally or distributing runs to the cloud. It is written in Python, the repository was created on November 16, 2022, and it has 8,392 stars.
There are points that could not be confirmed. The changes in v3.0.0 and the benchmarks backing its performance are not included in the provided material, so the concrete changes can only be known by checking the release notes directly. The README introduces the framework as an open-source project, but the license type could not be confirmed. The momentum metric also comes out low at 4.3 out of 10.