Benchmark RadarAI BENCHMARK PROFILE
IntentionNav
IntentionNav evaluates active object search from implicit human instructions in 176 Isaac Sim scenes. Episodes provide free-text intent, RGB-D observations, and pose, with the target object withheld. The benchmark includes 500 intents over 64 categories and four intent modes.
- Released
- 2026-05-22
- Readiness
- Paper only
- Primary field
- Robotics & Autonomous Systems
Why it matters
Addresses the gap in object navigation where agents must infer targets from indirect human intent, showing persistent bottlenecks in target selection and terminal localization.
Motivation
Existing object navigation benchmarks usually tell an embodied agent which object category to find, such as microwave or chair.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.