Cpu inference engine speeds up language-guided robot actions
vla.simd: Efficient CPU Inference for Language-Conditioned Manipulation
RoboticsArtificial Intelligence
Summary
Robots that respond to language usually need powerful graphics processors to work efficiently. The authors developed vla.simd, a software that helps CPUs run language-guided robot programs faster without losing accuracy. They also created a new robot policy called IMPACT that works well on low-power devices like the Raspberry Pi, producing many robot actions per second. Their approach was tested on several robot arms, showing that fast, language-driven robot control is possible on everyday CPUs.
What this means in practice
- •For industrial robot programmers: Deploy language-guided manipulation strategies on standard CPUs without requiring GPUs, improving cost-efficiency in manufacturing automation.
- •For embedded system developers: Implement fast language-conditioned action policies on resource-limited devices like Raspberry Pi for compact robotic platforms.
Authors
Khanh D. Nguyen, Hoang M. Truong, An T. Le
Abstract
Deploying language-conditioned manipulation without a dedicated GPU requires efficient inference and action chunks that cover the delay between policy queries. We present vla.simd, a CPU inference engine that combines shared SIMD micro-kernels, reusable computation, and target-specific optimization. We relate query latency and execution horizon to action availability under lagged and time-aligned execution, distinguishing action supply from feedback frequency. Across six policies and four CPUs, vla.simd achieves approximately $1.4\times$ median speedup over compiled PyTorch references while preserving fp32 numerical fidelity. We also introduce IMPACT, an ACT-based policy with cached text representations and language-modulated visual features. IMPACT is the only language-conditioned policy in our evaluated set that supplies at least 30 actions/s on the Raspberry Pi 5: after a 90 s thermal soak, it supplies 33.5 actions/s in fp32 and 81.2 with int8. Separate GPU evaluations yield $76.4\%$ mean success across four LIBERO suites without robot pretraining; instruction-shuffling tests demonstrate selection among familiar goals. Trials with IMPACT on an SO-101 arm and SmolVLA on a UR10e with a Robotiq gripper demonstrate CPU deployment on two robot embodiments.