- lio
- slam
- traversability
- vla
- vln
- ros2
- gazebo
- paper-review
- tech-note
•
•
•
•
•
•
•
•
-
π0: A Vision-Language-Action Flow Model for General Robot Control (arXiv 2024)
Review of π0, a robot foundation model that attaches a flow matching action expert to a pretrained VLM and trains on 10,000 hours of cross-embodiment data
-
π0.5: a Vision-Language-Action Model with Open-World Generalization (CoRL 2025)
Review of π0.5, which combines heterogeneous-data co-training with hierarchical subtask reasoning to perform long-horizon mobile manipulation in homes never seen during training
-
π*0.6: a VLA That Learns From Experience (arXiv 2025)
Review of π*0.6, which applies real-world RL to a flow-matching VLA with the RECAP recipe, learning from autonomous rollouts and human intervention corrections via advantage conditioning
-
NaVILA: Legged Robot Vision-Language-Action Model for Navigation (RSS 2025)
A two-level framework in which the VLA outputs only natural-language mid-level actions and a separate visual locomotion RL policy handles joint control, demonstrating legged-robot VLN from a single RGB camera
-
M2Mapping: Neural Surface Reconstruction and Rendering for LiDAR-Visual Systems (ICRA 2025)
A mapping framework that couples the LiDAR distance field and the camera radiance field through a single SDF-to-density transform, yielding watertight surfaces and extrapolated rendering together even on free-view trajectories
-
FSR-VLN: Fast and Slow Reasoning for Vision-Language Navigation (arXiv 2025)
A system that adds a view layer to a hierarchical multimodal scene graph and calls slow VLM reasoning only when fast CLIP matching fails, reaching 92% SR in 5.5 s on long-horizon indoor object search with a humanoid
-
Elevator-LIO: Robust LiDAR-Inertial Odometry for Multi-Floor Navigation under Elevator-Induced Non-Inertial Motion (arXiv 2026)
A multi-floor LIO that separates the robot's relative motion from the cabin's transport motion in the state inside a moving elevator and suppresses vertical drift with a ZUPT at the stop instant
-
Continual Learning for Traversability Prediction with Uncertainty-Aware Adaptation (RA-L 2025)
Continual traversability learning that recalls past experience with a CVAE instead of storing data, filtering recalled samples by uncertainty to reduce catastrophic forgetting