Jetson-PI: Boosting Onboard VLA Control Frequency to 6Hz, 8.6x Faster
机器之心 · wechat · 2026-07-25
Peking University, AIRS, and PrimeBot Research introduced Jetson-PI, a real-time Vision-Language-Action (VLA) control scheme for low-power edge devices. Through asynchronous inference, model scheduling, and engine optimization, it boosts control frequency on Jetson Orin from 0.7Hz to 6.06Hz without losing accuracy.
- Foresight-Aligned Asynchronous Correction: Instead of predicting future images, it trains a lightweight module (1% of parameters) to predict future environment representations, solving the perception-execution misalignment in asynchronous inference.
- Confidence-based Scheduling: Enables the robot to decide when to act on existing knowledge and when to re-invoke the VLM for fresh observations, reducing unnecessary compute.
- Engine Optimization: Developed Jetson-PI-Edge based on llama.cpp, leveraging CUDA graph reuse, persistent GPU caching, and flow matching unfolding to minimize overhead.
In real-world clothes folding tasks, this approach significantly reduced motion stalls and misalignment, pushing VLA control toward industrial deployment.
More from Embodied
- Humanoid walking is continuous correction — watch recovery, not locomotion, in demos — Olivier__OG · 2026-10-26
- CircuitHub unveils The Grid, a factory-scale robot making PCBs in the US twice as fast as Shenzhen — garrytan · 2026-10-26
- AI researchers guessed robots wouldn't outrun humans until 2034 — it already happened — ben_j_todd · 2026-10-26
- Shanghai AI Lab-led team releases 100-hour egocentric wearable dataset for humanoid training — ChuckDBrooks · 2026-10-26
- RealSense D585 Pro AI-Powered 3D Stereo Depth Camera Unboxed for Physical AI — chrismatthieu · 2026-10-26
- Physical world agent swarm demo shared on X — pduan · 2026-10-26