Neuromeka AI Lab

Welcome to Neuromeka AI Lab

  At Neuromeka AI Lab, we tackle industry-specific challenges by focusing on practical, field-ready AI solutions. By integrating advanced data-driven methodologies into industrial machines, we aim to deliver real-world impact and drive meaningful transformation in automation.


Open Positions & Contact

Please reach out to Joonho Lee (joonho.lee@neuromeka.com).

We are looking for creative and motivated team players!


Research

  Our current research focuses on equipping robots with new capabilities to address practical challenges in industrial settings. We begin by targeting specific industrial applications, with the broader goal of developing more generalizable robotic systems in the future.

  • Visuomotor Policy Learning: We develop robust visuomotor policies that bridge perception and control, enabling robots to interpret visual information and respond in real time. We explore learning-based control methods—such as imitation learning (IL) and reinforcement learning (RL)—both in simulation and on physical platforms.
  • Safe and Intuitive Interaction: We design systems that prioritize safety, reliability, and seamless human-robot collaboration in shared workspaces. By integrating intuitive control strategies and robust perception, we aim to make robots trusted partners in the workplace.

  • Mobile Manipulation: We aim to expand our robots’ workspace and adaptability by integrating them with mobile platforms, such as Neuromeka’s Moby and Moby Outdoor robots.

Publications

  • Lee, Joonho, et al. “Learning Fast, Tool-Aware Collision Avoidance for Collaborative Robots.” IEEE Robotics and Automation Letters (2025)

Ongoing Projects

 

Fast, Tool-aware Collision Avoidance for Cobots

  We develop fast and efficient collision avoidance techniques for collaborative robots (cobots) using point cloud data from sensors like Intel Realsense and 3D LiDAR. Our approach leverages constrained reinforcement learning (CRL) to ensure safe, tool-aware motion in dynamic environments while maintaining efficiency and responsiveness.

Publications:
  • Lee, Joonho, et al. “Learning Fast, Tool-Aware Collision Avoidance for Collaborative Robots.” IEEE Robotics and Automation Letters (2025)

Imitation Learning for Indy7 Robot

  We are actively exploring learning-based approaches that incorporate real-robot data and human demonstrations. In 2024, we showcased a public demonstration based on ACT (Action Chunking Transformer), showing the potential of imitation learning (IL) for novel skill acquisition.


Moby Animation

Mobile Manipulation

  We are preparing mobile manipulators, including Neuromeka’s Moby and Moby Outdoor. Our current focus is on building a GPU-accelerated navigation framework with real-time local mapping and path planning.

  Once our baselines are fully established, we will advance toward autonomous mobile manipulation in both structured and unstructured outdoor environments.