Can Astra Solve RoboMME without Breaking the Bank?
Can Astra Solve RoboMME without Breaking the Bank?
This research introduces a three-tier system for robotic manipulation that addresses memory challenges by using a vision-language planner to select subtasks from stored history, an action policy to execute them, and a monitor to verify completion. On the RoboMME benchmark, the system achieves a 79.13% success rate with an average of 3.63 calls to the large model per episode and 80.84 seconds per episode, outperforming the best non-oracle baseline by 34.62 points.
The authors present a three-tier system combining a vision-language planner, an action policy, and a monitor to achieve 79.13% success on the RoboMME benchmark with 3.63 average planner calls per episode.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
average_astra_calls
3.63
View original evidence
in 79.13% of episodes with an average of 3.63 Astra calls and 80.84 seconds per episode. This is 34.62 points above the best published non-oracle baseline and 4.95 points below GroundSG+Oracle, which gives the low-level policy ground-truth subgoals.
Open source S4average_episode_duration_seconds
80.84 seconds
View original evidence
in 79.13% of episodes with an average of 3.63 Astra calls and 80.84 seconds per episode. This is 34.62 points above the best published non-oracle baseline and 4.95 points below GroundSG+Oracle, which gives the low-level policy ground-truth subgoals.
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
in 79.13% of episodes with an average of 3.63 Astra calls and 80.84 seconds per episode. This is 34.62 points above the best published non-oracle baseline and 4.95 points below GroundSG+Oracle, which gives the low-level policy ground-truth subgoals.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org