WorldArena 2.0 Brings World Models to Real Robots
机器之心 · wechat · 2026-07-15
WorldArena 2.0 has been released alongside the IROS 2026 challenge, aiming to shift world model evaluation from simply assessing visual quality to testing real-world applicability.
Core Upgrades
- Track1: Evaluates video quality for complex OOD tasks, continuing to assess physical consistency, controllability, and 3D accuracy.
- Track2: Connects world models to an online reinforcement learning closed-loop to test their viability as policy training environments.
- Track3: Deploys on real robot platforms to evaluate WAM task planning and action execution under multi-embodiment, multi-modal, and distributed conditions.
Existing Experimental Results
- In tactile-intensive tasks, tactile feedback helps, but high prediction metrics do not guarantee long-horizon manipulation success.
- For example, on the InsertHDMI task, some world models achieve high prediction quality and success rates. However, for the LiftBottle task requiring continuous force control, all tested world models had a 0% success rate, whereas the ACT baseline reached 80%.
Event Details
- Opens on July 10
- Submission deadline on August 30
- Results announced on September 15
- Awards ceremony on September 27
- Paper, code, data, and evaluation framework are already open source
More from Embodied
- NVIDIA releases MotionBricks: real-time humanoid motion generation at 15,000 FPS with 2ms latency — _akhaliq · 2026-10-26
- Claude Code finds root backdoor in Honda Civic head unit, then debloats it — IndraVahan · 2026-10-26
- Temple health wearable opens pre-orders, users ask: why wear it before benefits are proven — abhishekcode42 · 2026-10-26
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- Luke Wroblewski spots it: iPhone Duo's corner radius mirrors the original iPhone — LukeW · 2026-10-26
- Humanoids shattered forecasts: half-marathon record in 2026, 100m faster than Bolt — ben_j_todd · 2026-10-26