TGTGInsightintelligence telegramLIVE / telegram public index
TGINSIGHT POST
Post #1608
@huggingface
Hugging Face
Contenuto del post
Contenuto
βHugging Face (Twitter) RT @LeRobotHF: π ππ’π π§ππ°π¬ ππ¨π« πππ ππ¨ππ¨ππ’ππ¬ ππππ«π§π’π§π ! ππππ¨ππ¨π π―0.4.0 π’π¬ ππππ! We're supercharging open-source robot learning with major updates to make it more powerful, scalable, and user-friendly. Highlights: * π¦ Dataset v3.0: A new chunked, streamable format built for massive (OXE-level) datasets. * π New Sim Support: Integrated LIBERO & Meta-World for diverse VLA & multi-task training and evaluation. * π οΈ Data Processors: A new modular pipeline to seamlessly connect any policy to any robot. * π§ SOTA Policies: We've integrated PI0 / PI0.5 & an Nvidia Gr00t N1.5 for open-world generalization. * β‘ Multi-GPU Training: Scale your experiments with Accelerate in just one command. * π Hardware Plugin System: A new era for hardware! Connect new robots to `lerobot` with a simple `pip install`. * π©βπ« New HF Robot Learning Course: We're launching a 100% open-source course to make robot learning accessible to all! This is a huge leap forward. Read the full announcement in the Release Blog: