Interlink
Our teleoperation platform that enables deployment today through remote control of robots from any location worldwide.
Interlink connects human operators with our robots, providing the foundation for robust and efficient real-world deployment. It serves two primary purposes: powering data collection for training our savant AI models, and providing a human-in-the-loop mechanism for handling edge cases.
Interlink supports data collection for training our savant AI models, which are designed to achieve human-level performance in task execution. Operators use Interlink to direct our robots in various applications, generating the datasets that facilitate the curation of our savant AI models.
Interlink provides a human-in-the-loop mechanism for handling edge cases where our AI models may become unsure of how to progress the task. Edge cases could involve complex scenarios, such as unpredictable object handling, or dynamic changes like sudden task state changes.
These scenarios are identified through our task progression model; if task progression stagnates or rapidly declines, Interlink notifies a teleoperator, who either suggests a suitable course of action to the model or assumes temporary control. Control then returns to the robot, and the intervention data is incorporated into model training, advancing toward greater autonomy.
Interlink comprises two main elements for effective and scalable teleoperation.
XO Exoskeleton
The operator interface features XO, our proprietary exoskeleton. It is designed for intuitive control and precision, with sensors that map human movements to robot actions. This allows for dexterous object manipulation.
Latency-Aware Control
A latency-aware control policy, trained via reinforcement learning, mitigates the effects of latency while generating robust and dexterous control. This ensures responsive performance over networks, maintaining safety and fluidity in all operations.
Interlink embodies Eden's approach to instilling human intelligence and performance within our AI models. By gathering high-quality data from real-world deployments today, we are accelerating the path to versatile, general-purpose robots tomorrow.