feat: add VLA 4-model benchmark leaderboard with GR00T N1.6

README.md

@@ -105,6 +105,22 @@ print(f"Danger zones: {len(danger)}")
 | GPU simulation | Isaac Lab | NVIDIA 2025 |
 | Grasp evaluation | Isaac Sim Grasping SDG | NVIDIA 2025 |
 
+## VLA Benchmark — 4-Model Leaderboard
+
+Four VLA models evaluated on RoboGate's 68-scenario adversarial suite. **All scored 0% SR** — including NVIDIA's official GR00T N1.6.
+
+| Model | Params | SR | Confidence | Failure Pattern |
+|-------|--------|-----|-----------|-----------------|
+| Scripted Controller | — | **100%** (68/68) | 76/100 | — |
+| **GR00T N1.6 (NVIDIA)** | 3B | 0% (0/68) | 1/100 | grasp_miss + collision |
+| OpenVLA (Stanford + TRI) | 7B | 0% (0/68) | 27/100 | grasp_miss dominant, 0 collision |
+| Octo-Base (UC Berkeley) | 93M | 0% (0/68) | 1/100 | grasp_miss 79%, collision 21% |
+| Octo-Small (UC Berkeley) | 27M | 0% (0/68) | 1/100 | grasp_miss 79.4%, collision 20.6% |
+
+Model size is not the bottleneck — even NVIDIA's flagship 3B model cannot bridge the training-deployment distribution gap.
+
+**Leaderboard:** [robogate.io/vla](https://robogate.io/vla) · **Paper:** [arXiv:2603.22126](https://arxiv.org/abs/2603.22126)
+
 ## Citation
 
 ```bibtex