Robot control training enters the minute-level era! Tsinghua AIR open-sources UniLab: trains humanoids in 3 minutes, speed increases 10x, can run on Mac
Researchers from Tsinghua University's AIR DISCOVER Lab have introduced UniLab, an open-source framework for robot reinforcement learning training. This new architecture utilizes a heterogeneous approach, offloading physics simulations to CPUs while leveraging GPUs for policy training, significantly reducing training times. UniLab achieves up to a 10x speedup in training humanoid robots and is compatible with various hardware, including Macs, breaking free from CUDA dependency. AI
IMPACT Accelerates robot training cycles, potentially enabling faster development and deployment of embodied AI systems.