VLA Handover: Failure-Prediction-Based Switching from On-Robot to Server VLA Models
Organizations: School of Engineering, Institute of Science Tokyo, Tokyo 152-8550, Japan
Abstract
Large vision-language-action (VLA) models generalize well but are difficult to run in real time on a robot, and small VLAs that run there fail more often. Remote inference over beyond-5G and 6G networks lets a robot use a large VLA on a server, but querying the server for every action chunk limits the number of robots per server. We propose VLA handover: the robot runs a small VLA (S-VLA) and switches to a large VLA (L-VLA) on a server only when the S-VLA is likely to fail. We formulate the switching as minimizing the request cost under a required success probability and, since this problem cannot be solved directly, propose a heuristic framework that retrains neither VLA. In simulation with SmolVLA and on three of the four LIBERO suites, VLA handover raised the success rate by 11-22 percentage points over the S-VLA alone and sent 49-77% fewer requests than server-only control at a 2-4 percentage-point lower success rate, and its completion time grew less with the network latency.
Figures & tables
| Scheme | Success | Req./ep. | Srv. frac. | Compl. [s] |
|---|---|---|---|---|
| Device-only | 83.8 [80.3, 86.8] | 0 | 0 | 7.36 |
| Server-only | 96.8 [94.9, 98.0] | 14.99 | 1.00 | 6.96 |
| Periodic | 96.4 [94.4, 97.7] | 15.00 | 0.96 | 6.62 |
| Periodic | 91.8 [89.1, 93.9] | 8.97 | 0.97 | 7.03 |
| Periodic | 86.8 [83.6, 89.5] | 6.59 | 0.97 | 6.82 |
| Ours: VLA handover | 94.4 [92.0, 96.1] | 3.40 | 0.18 | 8.51 |
| Suite | Scheme | Success | Req./ep. | Srv. frac. |
|---|---|---|---|---|
| Goal | Device-only | 76.3 | 0 | 0 |
| Server-only | 94.0 | 13.15 | 1.00 | |
| Periodic | 91.0 | 5.01 | 0.96 | |
| Ours: VLA handover | 91.0 | 5.32 | 0.29 | |
| – w/o arm reset | 77.7 | 7.66 | 0.41 | |
| – fixed step | 87.0 | 7.47 | 0.37 |
| H, success | H, failure | Req./ep. | |||
|---|---|---|---|---|---|
| Suite | Ep. | Req./ep. | Ep. | Req./ep. (share) | all w/o H-fail. |
| Object | 16.2% | 14.2 | 4.6% | 23.7 (32%) | 3.40 2.31 |
| Goal | 25.0% | 11.9 | 9.0% | 26.0 (44%) | 5.32 2.98 |
| Spatial | 41.7% | 9.5 | 10.3% | 24.4 (39%) | 6.46 3.94 |
| Long | 34.6% | 20.7 | 36.2% | 33.2 (63%) | 19.17 7.14 |
Efficient VLA Model Inference
Efficient vision-language-action (VLA) model inference reduces robot-control latency, memory use, and energy while preserving task performance.
See all 215 papers