COCO Matrix has introduced its latest robot model, the CO01, which leverages an innovative 'experience-driven learning' methodology. This model seamlessly integrates various forms of experiences, including live demonstrations, failed attempts, real-time instructions, and target images, directly into its operational context. This integration eliminates the necessity for retraining when undertaking new tasks, thereby significantly slashing the costs associated with robot adaptation and deployment.

Users can instruct robots through demonstrations and guidance, and the experiences garnered during these interactions can fuel continuous model refinement. Practical testing reveals that, under an identical pre-trained model, a mere minute of demonstration can elevate the average success rate for unfamiliar tasks to 52%. In stark contrast, achieving a comparable success rate through fine-tuning necessitates approximately 180 demonstrations and roughly 9 hours, culminating in a preparation time disparity of about 460 times.

A real-world robot that retains a record of its past failed attempts attains a cumulative success rate of 84% within five attempts, substantially surpassing the 56% success rate achieved by a robot without such historical retention. The CO01 model relies on conversational training to maintain experience associations, enabling it to modify behaviors based on past experiences without the need for weight updates. Its unified visual framework supports a diverse array of perception tasks, outperforming specialized models in 5 out of 8 benchmarks, thereby laying a robust perceptual groundwork for robots to comprehend demonstration content.

The CO01 model possesses the capability to learn from demonstrations, its own attempts, real-time guidance, and transfer skills across different robotic platforms. Moving forward, the team will focus on optimizing issues such as error accumulation in prolonged tasks and precision in experience selection, fostering the sharing and continual evolution of robotic experiences.