X-OP: Single XR Device Controls Any Robot via MPC Retargeting
Robots controlled across morphologies with 30% faster humanoid tasks and zero collisions.
X-OP (Cross-Morphology Whole-Body Teleoperation via MPC Retargeting) introduces a plug-and-play framework that allows a single extended reality (XR) device to control diverse robot morphologies (humanoid, mobile manipulator) without robot-specific retraining. At its core, a Model Predictive Control (MPC)-based motion retargeter jointly optimizes alignment with the operator's intent and the robot's dynamic feasibility, generating optimal commands for existing low-level controllers. Unlike prior methods that rely on expensive exoskeleton suits or multi-camera setups, X-OP uses a single headset and SLAM-based global pose feedback to mitigate long-term drift and a state synchronization method to handle noisy real-world measurements and contact sensitivity.
In simulation, X-OP outperforms baselines by over 30% lower completion time and 20% lower power consumption on humanoid tasks, and achieves zero collisions for mobile manipulators. Real-world experiments on both platforms confirm the method's effectiveness, enabling users to adjust teleoperation behavior based on preferences. This morphology-agnostic, scalable solution promises easier robot data collection for loco-manipulation tasks without prohibitive cost or complexity.
- Uses a single XR device and MPC retargeting to control humanoids and mobile manipulators without retraining.
- Reduces humanoid task completion time by 30% and power consumption by 20% vs baselines.
- Real-world validation on both platforms with SLAM-based drift correction and state synchronization.
Why It Matters
Enables affordable, scalable whole-body teleoperation across robot types, accelerating data collection for loco-manipulation tasks.