Psi-P0
Psi-P0 demonstrates the capability to decompose and plan complex tasks in open-ended environments. Both its task complexity handling and accuracy exceed the benchmarks set by concurrent works, including OpenAI's VPT and Nvidia's Minedojo.
TECHNICAL ARCHITECTURE
Psi-R0
PsiBot has recently introduced Psi-R0. The model features bimanual dexterous manipulation capabilities and employs mixed training of multiple skills to create reasoning-capable agents that can execute and close the loop on extended dexterous manipulation sequences.
-
-
Psi-C0
Psi-C0 combines human motion data and deep reinforcement learning: the upper level learns from human data to generate reference trajectories, while the lower level uses these trajectories to guide RL training.
PSIBOT G0
PsiBot has developed and mass-produced its first dexterous hand based on algorithmic needs. The data collection strategy combines simulated data gathered through teleoperation and real-world data captured via exoskeletons that accurately mirror human movements.
-
High Precision Positioning and Repeatability
-
Multi-Axis Freedom and Flexibility
-
Intelligent Path Planning and Adaptive Control
-
High Payload Capacity and Energy-Efficient Design
-
Modular Integration and Industry 4.0 Compatibility