PointMaze
A sparse-reward continuous-control navigation task in which the agent must reach a goal point in a maze to receive a reward, and the goal direction is not directly provided in the observation. It requires long-range memory and structured exploration capability, and is frequently used to test the effectiveness of long-context memory mechanisms.
Timeline (last 90 days)
On the long-context, sparse-reward PointMaze task, METTA exhibits more structured exploration behavior than Transformer-based baselines.
PointMaze is a sparse-reward continuous-control navigation task in which the agent must reach the goal point in a maze to receive a reward, and the observation does not directly give the goal direction.
Episodes that rely on frequent replanning to succeed are enriched with severe first-plan ranking failure cases in the PointMaze first-plan diagnosis.
All Facts (3)
On the long-context, sparse-reward PointMaze task, METTA exhibits more structured exploration behavior than Transformer-based baselines.
50%UnverifiedPointMaze is a sparse-reward continuous-control navigation task in which the agent must reach the goal point in a maze to receive a reward, and the observation does not directly give the goal direction.
50%UnverifiedEpisodes that rely on frequent replanning to succeed are enriched with severe first-plan ranking failure cases in the PointMaze first-plan diagnosis.
50%