Search papers, labs, and topics across Lattice.
and AI Institute
2
0
4
Achieving a 35% success rate in real-world robotic tasks using only synthetic data marks a groundbreaking advance in sim-to-real transfer for world-action models.
Ditch hand-tuned action penalties: this work uses an action Jacobian penalty and a Linear Policy Net to automatically learn smooth, realistic robot control policies without task-specific tuning.
