Benchmark RadarAI BENCHMARK PROFILE
CondVLN
CondVLN evaluates vision-language navigation agents on 11,500 generated conditional instructions across four simulators, using standard VLN metrics plus Branch Selection Accuracy and Conditional Success Rate.
- Released
- 2026-08-18
- Readiness
- Paper only
- Primary field
- Robotics & Autonomous Systems
Why it matters
CondVLN provides controlled diagnostic signals for conditional branching failures in VLN, showing that high success rates can mask incorrect branch execution and offering a reusable testbed for instruction following under conditions.
Motivation
Vision-language navigation agents are often evaluated on their ability to follow route-like instructions toward a fixed goal.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.