EagleVLA Selected as a CoRL 2026 Spotlight

Our CoRL 2026 paper EagleVLA (formerly Jetson-PI): Towards Onboard Real-Time Robot Control via Foresight-Aligned Asynchronous Inference has been selected as a spotlight, a distinction awarded to fewer than 5% of accepted papers. EagleVLA addresses the high latency and low control frequency of deploying vision-language-action models on low-power onboard devices by introducing a lightweight future-correction module that predicts future environment representations from committed actions, thereby aligning action predictions with the state of the environment at execution time; a confidence-based scheduler then adaptively balances vision-language model and action-expert invocations, while a llama.cpp-based inference engine accelerates deployment through CUDA graph reuse, GPU-resident intermediate buffering, and flow unrolling. On NVIDIA Jetson Orin, the system improves control frequency by 8.66× over naive PyTorch and 5.41× over vla.cpp, while achieving a 14.8% higher average success rate than VLASH on the LIBERO benchmark. This work is led by our PhD student Zebin Yang.