CAFE: Closed-Loop Autonomous Task Formulation and Execution without Instance-Specific Instructions
CAFE: Closed-Loop Autonomous Task Formulation and Execution without Instance-Specific Instructions
This research introduces CAFE, a system for scene-driven robot autonomy that selects and executes tasks within predefined categories like cleaning and storage without instance-specific goals. The architecture maintains online memory of objects and zones, using a rule-guided Trigger-LLM to admit formulation from evidence while Draft, Complete, and Audit modules propose goals, resolve details, and check dependencies. Evaluation on 12 AI2-THOR scenes compares three backbones and task-complexity tiers using coverage, constraint, and execution metrics, with results showing Tri-LLM achieving higher GCR scores than baselines. The paper includes controls on completion, auditing, zones, feedback, triggering, and prompt examples, alongside LimoCobot traces illustrating storage, cleanup, and recovery from failed grasps.
The study evaluates a closed-loop system that formulates and executes cleaning, safety, and organization tasks in AI2-THOR scenes without instance-specific instructions, reporting higher GCR scores than baselines across three backbones and task-complexity tiers.
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.
- Control
- Reported autonomousOpen source S1
Reported numbers
scenes evaluated
12
Reported trials
View original evidence
rigger-LLM admits formulation from accumulated evidence; Draft, Complete, and Audit propose goals, resolve execution details, and check dependencies. Outcomes update memory for correction and subsequent discovery. Evaluation in 12 AI2-THOR scenes compares three backbones and task-complexity tiers using coverage, constraint, and execution metrics. Tri-LLM records higher GCR scores than the evaluated baselines across t
Open source S3
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
CAFE: Closed-Loop Autonomous Task Formulation and Execution without Instance-Specific Instructions
Open source S1
rigger-LLM admits formulation from accumulated evidence; Draft, Complete, and Audit propose goals, resolve execution details, and check dependencies. Outcomes update memory for correction and subsequent discovery. Evaluation in 12 AI2-THOR scenes compares three backbones and task-complexity tiers using coverage, constraint, and execution metrics. Tri-LLM records higher GCR scores than the evaluated baselines across t
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org