SO-101

The SO-101 benchmark is a standardized real-world evaluation framework for assessing Vision-Language-Action models on a low-cost robotic platform, consisting of four manipulation tasks and incorporating structured failure analysis and recovery metrics.

Papers using SO-101 (3)

Benchmarking Vision-Language-Action Models on SO-101: Failure and Recovery Analysis2026Towards Accessible Physical AI: LoRA-Based Fine-Tuning of VLA Models for Real-World Robot Control2025ACG: Action Coherence Guidance for Flow-based Vision-Language-Action models2025