All roles

Robot Learning Research Engineer

Samseong-dong, Gangnam-gu, SeoulFull-time

About Us

Interact Robotics focuses on challenging contact-rich manipulation tasks that remain difficult for vision-centric robot AI. We develop dexterity, the ability to manipulate objects precisely with multiple fingers, and contact intelligence, the ability to adapt motion using fingertip tactile feedback and forces sensed through the wrist and arm.

We collaborate with leading university labs in Korea and abroad on haptics, robot control, and robot learning. We integrate the technologies we develop together into our models and robotic systems and validate them on real-world contact-rich tasks.

About the Role

We are looking for a Research Engineer to build on frontier robotics models, develop manipulation policies that use force and tactile feedback, and deploy them on real robots.

You will reproduce Vision–Language–Action (VLA) baselines on real-world tasks and explore how models can represent and use wrist force and torque and fingertip tactile signals. If you have experience with world models, you can help predict how candidate actions change contact states and use those predictions to select the next action. On tasks such as reorienting objects with a multi-fingered robotic hand or mating connectors, you will measure how force and tactile feedback affect task success, completion time, and contact forces.

What You’ll Do

  • Reproduce and fine-tune frontier robotics modelsEstablish baselines with models such as the π0 family and NVIDIA GR00T, then fine-tune them on in-house demonstration data for contact-rich tasks.
  • Learn and integrate force and tactile representationsAlign video, robot state, force/torque, and tactile signals sampled at different rates. Design encoders and multimodal fusion architectures that preserve brief changes in contact, then integrate the learned haptic representations into VLA policies or action-conditioned world models.
  • Build training and evaluation pipelinesPrepare demonstration data for training, compare vision-based baselines with models that also use force and tactile inputs under matched conditions, and evaluate generalization to new parts and environments.
  • Deploy policies on real robots and analyze failuresWork with the robot control and data collection teams to deploy policies on hardware. Use logs and repeated experiments to diagnose failed connections, slip, excessive contact forces, and delayed responses. Use those findings to guide data collection and model improvements.

Required Qualifications

  • Hands-on experience implementing, training, and evaluating models in at least one of the following areas: VLA or robotics foundation models, manipulation policies based on imitation or reinforcement learning, or action-conditioned world models or dynamics prediction
  • Proficiency in Python and either PyTorch or JAX, with the ability to preprocess data, train and evaluate models, and analyze failures
  • Experience validating manipulation policies or predictive models on real robots or in simulation and improving them based on experimental results
  • A PhD in robotics, computer science, AI, or a related field (including candidates expected to receive their degree by February 2027), or at least three years of industry R&D experience in one of these fields.

Preferred Qualifications

  • Experience fine-tuning or deploying robot policies such as the π0 family, GR00T, or Diffusion Policy on real robots
  • Experience working with time-series sensor data such as force/torque, tactile signals, or robot state, or learning multimodal representations
  • Experience with action-conditioned future-state prediction, planning with world model rollouts, or closed-loop control using high-frequency sensor feedback
  • Experience building robot demonstration datasets, training models across multiple GPUs, or evaluating generalization across tasks and robot platforms
  • Publications, open-source contributions, or real-robot projects in robot learning, multimodal models, or world models

How We Work

You will design experiments with researchers in haptics, robot control, and data collection, from deciding which demonstrations to collect to evaluating learned policies on hardware. What the robot actually does guides the next experiment. Can it detect when a part binds during insertion and adjust its pose? Can it respond in time when an object starts to slip?

Compensation and Work Environment

  • Top-tier compensationWe offer compensation at the top end of South Korea’s robotics and AI market, reflecting your R&D experience, technical expertise, scope of responsibility, and expected contribution.
  • High-performance computing resourcesWe provide GPU compute for large-scale model training and iterative experimentation.
  • State-of-the-art robotics equipmentCollect data and validate learned models on hardware using modern humanoid robots, high-DoF robotic hands, and force and tactile sensors.
  • A workspace designed for researchWe are setting up a dedicated space in Samseong-dong, Seoul, for focused individual work, team collaboration, and hands-on robotics experiments.

What to Submit

Please share your résumé and one or two projects in which you played a leading role. You can share a paper, code, technical documentation, or a video of a real-robot demo. Briefly describe the problem you tackled, what you implemented yourself, and how you validated the results.

Hiring Process

  1. 01Application review
  2. 02Technical interview
  3. 03Final interview with the CEO