PhAIL and Nebius Launch Robot Policy Competition

The PhAIL team and Nebius launch a robot policy competition: participants train policies and package them as Docker images for submission, which are first scored in two simulation environments, MolmoSpaces and RoboLab, with RoboLab enabled starting November 16. In December, the top ten participants will run the same images on real Franka robots, with tasks similar to the DROID benchmark but using different datasets. Policies must fit within 24 GB of VRAM (equivalent to one 4090 GPU) so that participants can test locally. The prize pool is $23,500 worth of Nebius GPU and evaluation credits, including $1,000 awarded for best open-source writing. A follow-up report refers to the organizers as Positronic Robotics and Nebius, describing it as an international physical AI competition for developers and research teams, running until December 2026, with the prize described as over $20,000 in Nebius GPU compute and Positronic evaluation credits.

AI 根据报道生成 · 1 days ago更新