so3-sync-ring-6

SO(3) attitude synchronisation · solved 14 of 14 agent attempts pass

The task. Textbook-adjacent (consensus on a Lie group) but the agent only sees relative rotations and must respect the rate limit; the starter applies zero control.

20 verifier-only held-out seeds. A stock agent (cursor-cli, claude-sonnet-5-thinking-high) solved it in one edit on 2026-10-21.

Gate.

Axes: topology:ring, n_agents:6, observation:relative-only, dynamics:kinematic-so3

Each video re-runs the attempt's final controller through the task's own evaluator on a public seed (never the grader's held-out ones). The table under each video is the official grade on the held-out seeds.

Best attemptGemini (cursor-cli (pre-clock), 15 min) · base

Graded on the held-out seeds: passes every gate on the grader's held-out seeds.

criterion	graded	needed
base:success	1.000	≥ 0.900

How the agent got there: 15 min · 55 tool calls · 6 edit-tool edits · 11 runs of its own Python (tests, experiments, rewrites) · no tuner runs · official verifier 5× (success rates in order: 0.00, 1.00, 1.00, 1.00, 1.00).

The agent's final message

The SO(3) attitude synchronization task has been completed and verified. ### Summary - **Controller**: `/app/work/controller.py` implements a distributed Lie-algebra consensus law $\omega_i = k \sum_{j \in \mathcal{N}_i} A_{ij} \mathrm{Log}(R_i^T R_j)$ with gain $k = 8.5$. - **Verification Results**: - **Success rate**: 100% (`solved=True` across visible and held-out seeds). - **Final geodesic error**: $< 10^{-16}$ rad (well below the $0.02$ rad threshold). - **Mean total reward**: $-1.65$ (up from $-208.52$ with zero control). - Scratch and temporary test files have been cleaned up, leaving `/app/work/controller.py` ready for evaluation.

Controller family: consensus protocol

Each agent computes the log map of its relative rotation to each neighbor, sums these error vectors weighted by adjacency, and multiplies by a constant gain of 8.5 to produce its angular velocity command.

Shown: public seed 0 (base); this episode passes the default rung. This episode: final_edge_error 3.08e-17, final_pairwise_error 3.19e-17, total_reward -2.1.

controller.py · trial so3-sync-ring-6__8UdP2KA