LIBERO
LIBERO is a widely used simulation benchmark for robot manipulation and mobile manipulation research. It serves as a standard environment for evaluating the performance of Vision-Language-Action (VLA) models, World Action Models (WAMs), and various robot learning algorithms.
Several studies published in September 2026 utilized LIBERO as a key evaluation metric. The MM-ABC foundation model achieved a 99.1% success rate on the benchmark, while MoWAM and Rolling-WAM demonstrated manipulation performance competitive with existing methods. SafeLoop reduced hazard cases by approximately 70% across 24 LIBERO tasks while maintaining task success rates, and Action-JND improved compression reliability. WAA reached a 75.6% average success rate on the LIBERO-Pro extension, and GeoAAC improved simulated task success by up to 8.7 percentage points.
Key facts· every fact cites a source
| MM-ABC Success Rate | 99.1%source: learning note — MM-ABC: A Foundation Model for Mobile Manipulation |
|---|---|
| SafeLoop Hazard Reduction | approximately 70%source: learning note — SafeLoop Wraps Robot AI Models With a Built-In Safety Net |
| WAA LIBERO-Pro Success Rate | 75.6%source: learning note — New Multi-Agent Framework Lets Vision-Language Models Rehear… |
| GeoAAC Performance Gain | up to 8.7 percentage pointssource: learning note — GeoAAC Teaches Robot Policies When to Look Further Ahead |
Learning notes behind this entry7
- Sep 29 · paper · arXiv cs.ROMM-ABC: A Foundation Model for Mobile Manipulation
- Sep 25 · paper · arXiv cs.RORolling-WAM Speeds Robot Replanning With Staggered Denoising
- Sep 25 · paper · arXiv cs.RONew Multi-Agent Framework Lets Vision-Language Models Rehearse Robot Moves
- Sep 22 · paper · arXiv cs.ROSafeLoop Wraps Robot AI Models With a Built-In Safety Net
- Sep 18 · paper · arXiv cs.ROGeoAAC Teaches Robot Policies When to Look Further Ahead
- Sep 18 · paper · arXiv cs.ROMoWAM Swaps Future Video for Motion Prediction in Robot AI
- Aug 22 · paper · arXiv cs.RONew "Action-JND" Method Trims Robot AI Models Without Changing Behavior
Revision history2
·daily rewrite·local AI · qwen3.8 27b
Rewrote the summary from the remaining sources after 1 source note was withdrawn.
from: MM-ABC: A Foundation Model for Mobile Manipulati…, MoWAM Swaps Future Video for Motion Prediction i…, GeoAAC Teaches Robot Policies When to Look Furth… +4 more
·first version·local AI · qwen3.8 27b
First summary, written from 7 learning notes.
from: MoWAM Swaps Future Video for Motion Prediction i…, GeoAAC Teaches Robot Policies When to Look Furth…, SafeLoop Wraps Robot AI Models With a Built-In S… +3 more
next scheduled rewrite (if new notes arrived):
A local AI model wrote this entry from the learning notes. If something is wrong or missing, tell us — the AI reads and answers your report first, and the operator reviews it; any fix shows up in the entry's revision history.