Owned research — Hyperion

Reachy-LeRobot

A feasibility study: running a 4B-parameter vision-language-action policy (π0.5) on a single AMD Strix Halo workstation, with a Reachy Mini robot and SO-101 arms.

What we measured

  • π0.5 full fine-tune — peak memory

    Measured

    68.71 GB

  • Real fine-tune (LIBERO, language-conditioned) — loss

    Measured

    1.08 → 0.13 (300 steps)

  • Simulation success (LIBERO-Spatial) — base → fine-tuned

    Simulation

    0% → 96%

  • On-robot cognitive benchmark (12 scenes) — perceive · understand · interpret

    On-robot, dry-run

    12 / 12

  • bf16 throughput vs FP32 (ACT)

    Measured

    2.3× – 14.3×

Not yet measured

  • On-robot task success rate — the arm loop is dry-run (no physical motion yet).
  • NPU (XDNA2) execution — the INT8 toolchain validates, but no DPU subgraph compiled; iGPU/CPU fallback only.
  • Convergence (the fine-tune stops at 300 steps), energy, and multi-task generalisation.
Discuss embodied-AI work
Reachy-LeRobot — embodied-AI feasibility study | Hyperion