VLAN: leaderboard
Vision-language models navigate for different users and embodiments. Task success and accessibility-aware outcomes remain separate. Source contradictions in the drone results are retained for correction rather than clipped or substituted.
Source: arxiv.org.
Benchmark results
Results are published separately for each benchmark or evaluation setting.
- VLAN - (a) Humanoid - Instruction Agreement
- VLAN - (a) Humanoid - Semantic Success
- VLAN - (b) BVI - Instruction Agreement
- VLAN - (b) BVI - Semantic Success
- VLAN - (c) Wheelchair - Instruction Agreement
- VLAN - (c) Wheelchair - Semantic Success
- VLAN - Matched 200 Trajectories - BVI
- VLAN - Matched 200 Trajectories - Drone
- VLAN - Matched 200 Trajectories - Humanoid
- VLAN - Matched 200 Trajectories - Quadruped
- VLAN - Matched 200 Trajectories - Wheelchair
- VLAN - Quadruped Full Split - Collision Free
- VLAN - Quadruped Full Split - Goal Reached
- VLAN - Quadruped Full Split - SPL
- VLAN - Quadruped Full Split - Success
- VLAN - Real World 30 Trajectories - πππππ β
- VLAN - Real World 30 Trajectories - ππactππ―π β
- VLAN - Real World 30 Trajectories - ππlanππ―π β
Interactive version: theaggregate.ai/benchmark?slug=vlan Β· How It Works Β· Data refreshed daily, snapshot 2026-11-17.