rMuscle caches visual-token outputs and neuron activation patterns across similar repeated vision-language-action executions. On RTX 4090, Jetson Thor, LIBERO, RoboTwin and physical manipulation tasks, the authors report 1.29x-1.42x faster inference while preserving the original real-robot success rates. The benefit targets repetitive factory-style work rather than general robot behavior.
rMuscle reuses internal VLA state across repetitive robot tasks
2026-09-17Frontier watchWATCH
rMuscle caches visual-token outputs and neuron activation patterns across similar repeated vision-language-action executions. On RTX 4090, Jetson Thor, LIBERO, RoboTwin and physical manipulation tasks, the authors report 1.29x-1.42x faster inference while preserving the original real-robot success rates. The benefit targets repetitive factory-style work rather than general robot behavior.
Imported from the existing research archive. Migration preserves the recorded judgment and is not a fresh verification.
What this means for you
For robotics teams with repeated trajectories, internal-state reuse can lower latency without changing the VLA model. The workload assumptions are narrow and no code is linked. Reproduce on the actual camera, task variation and safety envelope before planning deployment.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Caches | visual-token Context Cache and activation-pattern Action Cache |
|---|---|
| Hardware | RTX 4090 and Jetson Thor |
| Benchmarks | LIBERO, RoboTwin and physical manipulation |
| Speedup | 1.29x-1.42x |
| Code | not linked |
Technical impact 3
Engineering value 3
Adoption signal 1
Maturity 1
Verification cost 5
Risk 3