Robotics Engineer (Teleoperation / UMI)

About the role

The AI Robot Association (AIRoA) is a national project selected by METI and NEDO to collect large-scale real-world robot data and develop generative AI foundation models for robotics. This position works on the teleoperation platform for humanoid robots, mobile manipulators and service robots, within the Development Division. The work is based in Tokyo.

Key responsibilities

• Design and implement teleoperation systems for humanoid, mobile manipulator and service robots, continuously improving operability, stability, latency and recovery performance.
• Establish operation logs, sensor logs, failure classifications, reproduction procedures and data collection workflows to improve the quality of human demonstrations used for imitation learning and VLA evaluation.
• Work with the Autonomy, VLA, Simulation, Integration and Hardware teams so that teleoperation, control, testing and real-robot evaluation run as an integrated system.
• Analyze logs on low-latency communication, control cycles, sensor synchronization, abnormal states and recovery, and own the reproduction, fixing and verification of issues on real robots.

Required profile

• Hands-on experience controlling or debugging physical robotic systems such as manipulators, humanoid, industrial or service robots.
• Experience in teleoperation, including VR, haptics, force feedback, leader-follower systems, motion capture, puppeteering-based control or real-time retargeting.
• Ability to troubleshoot hard-to-reproduce issues on physical systems through iterative logging, hypothesis development, reproduction testing, fixes and validation.

Required skills

• Building robotic systems with ROS or ROS 2, including system integration and system-level analysis.
• Preferred: implementation of robot control, real-time systems, communications, log analysis or related tools in C or Python.
• Preferred: experience with learning-based control (imitation learning, reinforcement learning, hybrid MPC + learning, safety-constrained learning) and end-to-end data collection workflows.