RLinf

Headlines · 3

1. 2d
   Tsinghua, Infinigence AI and AgiBot propose Collective Experience Learning for 96 robots

   Tsinghua University, Infinigence AI (无问芯穹) and AgiBot jointly released the Collective Experience Learning (CEL) paradigm, letting robots learn from peers' real task trajectories instead of relying only on human demonstrations. CEL pairs the FERT failure-guided experience-retrieval fine-tuning algorithm with the FATS fault-tolerant asynchronous training system, built on the open-source RLinf framework to coordinate up to 96 real robots. According to the project team, CEL surpassed the average performance of six human remote operators across five real tasks (flower arranging, folding clothes, phone charging, stool assembly and phone packaging), and on phone packaging the success rate rose from 86.7% with 16 training robots to 98.3% with 96.

   量子位（原生 RSS） · 🔥 6
2. 4d
   FluxVLA Engine V2 expands the embodied AI development stack

   LimX Dynamics updated the open-source FluxVLA Engine V2 technical report, adding online reinforcement-learning post-training, broader simulation evaluations, inference optimization, and deployment testing across ALOHA, Rebot, dual-arm Franka, and LimX Oli. The project reports that SmolVLA improved by 10.0 percentage points on LIBERO-10 and 15.4 points on ten RoboTwin Clean tasks after task-specific PPO post-training, while optimized inference paths across eight model-device combinations ran 2.31 to 9.64 times faster. The stack also supports multiple VLA architectures, including OpenVLA, π0.5, DiT4DiT, GR00T, and API-only policies, through shared data, training, evaluation, inference, and deployment workflows.

   雷峰网 AI · 🔥 0
3. Oct 8
   Zhengxing Innovation launches 24/7 Physical AI retail solution with H1 and C1 robots

   At the 22nd Asia Pacific Retailers Convention and Exhibition (APRCE 2026), Zhengxing Innovation unveiled a Physical AI solution for open retail settings built around human-robot collaboration: the general-purpose bipedal humanoid H1, the wheeled-arm handling and shelf-restocking robot C1 designed for narrow aisles, and M1, a unified robot management platform handling task allocation, status monitoring and human takeover. The company says its in-house lightweight world-action model SLM-0.5 reaches 98.6% average task success on the LIBERO benchmark with 24x faster inference than mainstream approaches, and that its RLinf-based reinforcement-learning post-training system STEAM achieves 95% success on a designated-product restocking task. Commercial service is planned for 2027 via direct purchase or RaaS subscription, with joint testing alongside global retail chains.

   量子位（原生 RSS） · 🔥 0

Experience and discussion from the community

Share my RLinf experienceAsk about RLinf