GoKawiilGeneral Instinct open-sources InstinctFlash, a fast serving engine for robotics AI models
General Instinct released InstinctFlash, an open-source inference framework that speeds up robotics 'world-action' models like LingBot-VA and pi05 on Nvidia's Jetson Thor edge computer, RTX 4090 and RTX 5090 GPUs. The company reports benchmarks showing up to a 33.78x speedup over standard PyTorch inference, achieved through FP8 quantization, reduced sampling steps, and custom acceleration kernels, with no measured drop in task performance on real robots.
GoKawiil's interpretation of the reporting above, not reported fact.
Robotics models that predict actions from vision and language typically run too slowly for real-time control on embedded hardware, forcing tradeoffs between model size and responsiveness. A framework that cuts inference latency by multiple times on compact edge devices like Jetson Thor could let robot developers deploy larger, more capable models without waiting for next-generation chips, accelerating progress in physical AI applications such as manipulation and navigation.
- InstinctFlash claims up to 33.78x faster inference for robotics action models compared to PyTorch baselines.
- The framework supports Nvidia Jetson Thor plus RTX 4090 and RTX 5090 GPUs with a shared Runtime API.
- It is fully open-sourced, covering eight robotics model families with Python and WebSocket serving options.
NVIDIA Jetson AGX Thor Developer Kit — This is the exact hardware InstinctFlash is benchmarked on, so it's the natural choice if you want to actually run these accelerated robotics models yourself. It gives robotics developers a compact edge platform for testing real-time inference workloads like the ones described in the article.
See NVIDIA Jetson AGX Thor Developer Kit on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: github.com, 2026-09-22
Published there as: “Show HN: InstinctFlash – Run 5B world-action models in real time on Jetson Thor”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.