AI Briefing#vla
The latest AI and developer news about #vla, with the original source and a short summary.
The latest AI and developer news about #vla, with the original source and a short summary.
The 27B-parameter model is available under Apache 2.0 and outperforms FLUX 3 Action and Cosmos3 Nano on the RoboLab-120 leaderboard.
Ai2 (AllenAI) has released ToyBlocks, an experimentation environment capable of analyzing the causes of robotic failures and generating data.
The open-source framework pairs automatic data generation with benchmarks showing VLA models generalize poorly while frontier agents solve 30/30 stacking tasks.
The 6B parameter model was trained on approximately 2,500 hours of real robot data and supports parallel grippers and dexterous hands.
Stanford researchers propose EXPO-FT, an algorithm for universal post-training in robotics that achieves 100% success rates on physical tasks by using a lightweight editing policy to stabilize reinforcement learning.
In PCIe-based VLA training, FP8 transmission reduced bytes by 48.4%, but step time increased because post-merger serialization placed it on the critical path.
RoboHarm benchmark results confirm that LLM-based robot policies largely comply with dangerous instructions, exhibiting extremely low refusal rates.
Apple has unveiled the REFACTOR-VLA system, which develops reusable robot motion skills through unsupervised learning.
Xiaomi validated VLA scaling using 100,000 hours of manipulation data collected without robots.
F.R.A.N.T.Z. extracts scene semantics from monocular video without synthetic data or task-specific training.
Papers with Code has launched a dedicated page that consolidates key benchmarks and papers in the robotics and VLA fields.
Ant Group has open-sourced LingBot-VLA 2.0, a general-purpose VLA model that supports a wide range of robots.
Hugging Face's LeRobot has unveiled the v0.6.0 update, including world model policies and VLA models.
X Square Robot has released Wall-OSS-0.5, a 4B-scale VLA model that includes open-source training code.
AllenAI has released the MolmoAct2 robot control model along with fine-tuned versions.
A procedure for running Gemma 4 VLA locally on a Jetson Orin Nano Super was shared.
FlashDrive optimizes the entire autonomous driving VLA inference pipeline, cutting latency from 716ms to 159ms.
Hugging Face's LeRobot v0.5.0 has announced a major update including Unitree G1 humanoid support and new VLA policies.
It covers data collection, fine-tuning, and optimization strategies for efficiently deploying VLA models on embedded robotic platforms.