CoBoVLA

Small VLA (250.3M deployed parameters) that maps language to discrete codes before a flow-matching actor produces actions, for robustness to paraphrased instructions. Results are self-reported on LIBERO-Goal only: 91.4% vs 82.8% for SmolVLA under the same protocol, which is not a LIBERO average. No code or weights are linked.

UC San Diego & Carnegie Mellon University · Nov 17, 2026 · VLAReport an error
250MParameters
0Benchmarks reported
Nov 17, 2026Released
UnknownWeights

Benchmark results0

No results on the tracked benchmarks yet.

Not reported on 15 benchmarks
Wrong or missing number? Report an error