Benchmark RadarSIS-Bench
SIS-Bench evaluates embodied spatial intelligence in UAV scenarios across two dimensions (spatial cognition and self-awareness) and three cognitive levels (perception, memory, reasoning), with 4,856 QA pairs from 1,646 real-world UAV videos.
- Released
- 2026-07-14
- Readiness
- Runnable
- Primary field
- Robotics & Autonomous Systems
Why it matters
Existing UAV benchmarks are environment-centric, leaving agent self-awareness implicit. SIS-Bench provides a unified self-in-space evaluation protocol, revealing imbalances between spatial and self-related cognition and offering practical value for developing UAV embodied models.
Motivation
Autonomous UAV systems increasingly rely on multimodal large language models (MLLMs) to operate in complex real-world environments.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.