Resolution criteria

This market resolves to the top score (Q-score) achieved by the 1st place team on the Held-out Test set during the 2026 BEHAVIOR Challenge organized by the Stanford Vision and Learning Lab (SVL).

• Source of Truth: The official leaderboard hosted on the BEHAVIOR-1K 2026 Challenge Leaderboard Hugging Face Space or the official BEHAVIOR challenge website.

• Metric: The primary ranking metric "Q-score" (average task success rate with BDDL partial credit) on the "Held-out Test" set.

• Resolution Value: previous year used 4 decimal points. Will select nearest interval if fewer or used and will round if more are used

• Timing: The market will resolve following the official announcement of the winners (scheduled for November 4, 2026). If the announcement or final leaderboard release is delayed, resolution will occur once official results are finalized by the organizers.

Background

The BEHAVIOR Challenge, hosted by the Stanford Vision and Learning Lab, is a benchmark designed to evaluate embodied AI agents on long-horizon mobile manipulation and navigation activities in realistic simulated household environments. The 2026 edition of the challenge expands the benchmark to 100 full-length household tasks in the BEHAVIOR-1K environment.

Rather than utilizing a binary success rate, organizers rank policies using a "Q-score". This metric tracks the fraction of completed symbolic sub-goals to reward partial credit for meaningful progress. In the 2025 challenge, the winning entry by the "Robot Learning Collective" achieved a peak Q-score of 0.2599 (25.99%) on the held-out test set.