A policy agent specifies local intent per action, and a critic evaluates outcomes and failure modes against that intent
Explore technology by industry and role
Newest additions first. Explore past and current notes by industry and role.
All notes
Newest additions first. Explore past and current notes by industry and role.
A structured demonstration strategy that progressively increases task complexity enables more efficient and stable learning in VLA models.
By leveraging the unexecuted tail of the previous action chunk, SEAM enables smooth transitions between action chunks without policy retraining.
Failure-driven analysis reveals intrinsic initialization as the dominant failure factor in fisheye calibration
Physical slip tests on Franka Panda robots in simulation provide validated grasp quality scores
Proposes a method to learn visual task spaces based on visual state changes, especially for cleaning tasks
An open-weight 27B reasoning model is trained to natively use real robots through simulation-based reinforcement learning.
A modular platform based on the 2018 KIA Soul EV is introduced for autonomous driving research
Custom instructions enable accelerator orchestration and reduced load/store latency for AI workloads.
A single VLA policy handles all 100 long-horizon household tasks efficiently.
A task-aware grasp selection framework that optimizes grasp choice based on goal conditions is proposed
Generalized AI models outperform specialized policies in direct manipulation control
Ferromagnetic semiconductor nanotubes with room-temperature Curie points are predicted, especially in CrS₂ and CrTe₂ zigzag structures exceeding 300 K
The D&A method separates progress estimation into a descriptive phase and an assessment phase, improving accuracy in zero-shot robotic task progress prediction.
A unified control and learning pipeline that aligns with human demonstration and interaction.