pi0.5

Headlines · 3

1. 12h
   NVIDIA Launches Cosmos Tools for Physical-AI World Models

   NVIDIA introduced Cosmos as a platform of world foundation models and development tools for physical AI, with Cosmos Framework covering training, multimodal data processing, optimization, and inference. The accompanying workflow covers synchronized camera and LiDAR simulation, GPU-accelerated deployment through NIM microservices, and support for models including Alpamayo, GR00T, Qwen, and pi0.5; these capabilities are described in NVIDIA Technical Blog.

   NVIDIA Technical Blog · 🔥 7
2. 5d
   Geek+ introduces Gravity Control for robot runtime error correction

   Geek+ recently introduced Gravity Control under its Gravity embodied-intelligence framework. The report says the method uses causal interventions on internal model features so natural-language prompts such as “grab the bread deeper” can correct local action deviations without updating the original policy weights or adding demonstration data. In reported physical-robot tests on out-of-distribution manipulation tasks, overall success rose from 17.5% to 87.5%, while the sideways-bread task improved from 5% to 95%. The method is intended to correct execution deviations when the robot already possesses the relevant skill, rather than teach it a new task.

   新智元 · 🔥 0
3. Oct 6
   Long-WAM scales world-action model context for real-time robot control

   Long-WAM is a model-system framework for scaling the context of causal world-action models under real-time robot-control constraints; its authors report that longer histories only pay off when the video foundation is pretrained autoregressively. On RoboCasa GR-1, raising context from 0.0 to 19.2 seconds lifted success from 63.3% to 78.7%, while a bidirectionally pretrained initialization showed no net gain, and robot-domain autoregressive pretraining further raised peak success on GR-1 and LIBERO-Long. The paper also reports deployment on RTX 5090, DGX Spark and Jetson AGX Thor, with each action chunk including future-video latent prediction taking 107.4 ms on RTX 5090, plus real-time long-horizon manipulation on Unitree G1 and YAM reaching 95% success on dynamic cup stacking, where it says Pi0.5 and Fast-WAM succeeded in none of 20 trials.

   Hugging Face · Papers · 🔥 0

Experience and discussion from the community

Share my pi0.5 experienceAsk about pi0.5