September 4, 2025 by Ozan Karaali

ICAT 2025: Using Visual Language Models to Control Bionic Hands

We presented our ongoing work at ICAT 2025 on leveraging visual-language models to improve object perception and grasp inference for bionic hands. The approach pairs an onboard camera with a lightweight client that streams context to an edge model, returning interpretable grasp suggestions.

In early evaluations on everyday objects, the pipeline produced stable grasp proposals and helpful natural‑language rationales, which we use to guide control strategies and user interaction. Next steps focus on latency‑aware execution and user studies in real‑world tasks.