Psi R0 - Lingchu Intelligence's end-to-end embodied model

What is Psi R0?

Psi R0 is Lingchu Intelligence's first end-to-end embodied model based on reinforcement learning. It supports complex operations through the collaboration of two dexterous hands, and can chain and train multiple skills to generate an intelligent agent with reasoning capabilities, completing and closing the loop on long-range dexterous operation tasks. Psi R0 can achieve cross-object and cross-scene generalization, and has strong generalization ability and high robustness.

Main functions of Psi R0

• Two dexterous hands working togetherThe Psi R0 supports complex operations performed by two dexterous hands in collaboration, enabling it to complete long-range dexterous tasks with multiple steps.
• Multi-skill integrated trainingThe model can chain and train multiple skills together to generate an intelligent agent with reasoning ability, which can complete and close the loop of long-range dexterous operation tasks.
• Cross-item and cross-scenario generalizationPsi R0 can achieve cross-item and cross-scene generalization, and has strong generalization ability and high robustness.
• Training based on simulation dataPsi R0 uses massive amounts of simulation data to train a two-handed intelligent agent, and connects multiple skills through a two-way training framework, making it the first in the industry to complete long-range tasks in open environments.
• Solving the problem of reward function designThis skill training framework abstracts key information from the spatiotemporal trajectory of objects to construct a general objective function, thus solving the problem of designing a reward function.
• Post-training phase optimizationIn the post-training phase, alignment with a small amount of high-quality real-machine data further improves the success rate of long-term tasks.
• Autonomous skill switching abilityThe transfer feasibility function in the bidirectional training framework can fine-tune skills to improve the success rate and generalization of cascading, while giving the model the ability to switch skills autonomously and quickly adjust strategies when encountering operational failures to ensure a high success rate.

Technical principles of Psi R0

• Reinforcement Learning (RL)Psi R0 is an end-to-end embodied model based on reinforcement learning, which uses massive amounts of simulation data to train an intelligent agent that operates with both hands.
• Skills training frameworkThis skill training framework abstracts key information from the spatiotemporal trajectory of objects to construct a general objective function, solving the problem of the difficulty in designing reward functions.

Application scenarios of Psi R0

• e-commerce scenariosThe Psi R0 can be used in the e-commerce industry for product packaging operations, involving multiple operations such as picking up, scanning, placing, and tying plastic bags for tens of thousands of items. The Psi R0 can smoothly complete this series of actions using two dexterous hands, replacing an entire workstation on-site.
• Factory production line assemblyIn manufacturing, Psi R0 can be used for assembly work on factory production lines to complete complex long-distance tasks, such as picking up, assembling and placing parts.
• Service industry picking and packingThe Psi R0 is also suitable for picking and packing tasks in the service industry, and can handle long-range tasks such as grabbing, scanning, and placing.
• Home cleaning and tidyingThe Psi R0 can also be used for cleaning and tidying in the home environment, handling everyday household chores.