Benchmark RadarAI BENCHMARK PROFILE
SceneActBench
A benchmark for visually conditioned action on complete multi-object 3D scenes across five tasks under a unified agent-environment loop, scored against hidden geometric ground truth.
- Released
- 2026-07-24
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Addresses the under-evaluation of agent action on complete 3D scenes, providing a comparative basis for VLM agents acting on 3D environments.
Motivation
Vision-language model (VLM) agents increasingly use tools to act on 3D scenes rather than only describe them.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.