The AI Research team at General Motors is looking to solve manipulation, planning, and simulation challenges at an industrial scale by developing end-to-end AI systems that enable dexterous manipulation, autonomous behaviors, and multimodal understanding on physical robotic platforms.
Requirements
- Proven experience in building and deploying ML models on robotic systems—including training, evaluation, and integration with real or simulated platforms.
- Deep understanding of modern AI architectures (e.g., Transformers, VLMs/VLAs, diffusion models, CNNs) and hands-on experience training models at scale.
- Strong implementation ability in PyTorch, including writing custom modules, batching, debugging, and performance/efficiency considerations.
- Practical experience with ROS/ROS2 or robotics middleware and integrating learning components into manipulation or motion-control workflows.
- Experience developing robot learning systems for manipulation, motion planning, or autonomous behaviors (e.g., diffusion policies, ACT, behavioral cloning, offline RL).
- Hands-on expertise with robotics perception, including 3D understanding, depth/RGB fusion, multimodal grounding, or force/torque sensing.
- Familiarity with simulation environments such as Isaac Sim, Mujoco, Gazebo, or PyBullet, and demonstrated experience with sim-to-real transfer strategies.
Responsibilities
- Design and implement advanced robot learning architectures (e.g., diffusion policies, ACT, VLM/VLA agents, imitation learning) to support manipulation, path planning, and autonomous execution.
- Build end-to-end model training pipelines for robotics applications, integrating multi-modal sensor data such as RGB, depth, force/torque, LiDAR, and proprioceptive signals.
- Develop scalable policy inference and control loops, pairing high-level perception with motion planning and on-robot execution.
- Apply or extend large-scale architectures—LLMs, VLMs, VLAs, diffusion models—to embodied tasks, sim-to-real adaptation, and grounding.
- Collaborate with cross-functional teams to translate research prototypes into deployable robotics software, ensuring robustness, efficiency, and safety.
- Design data collection, demonstration strategies, and simulation frameworks to support offline training, imitation learning, and hardware validation.
- Stay current with state-of-the-art advancements in embodied AI, robot learning, and manipulation, and share findings through internal research discussions and presentations.
Other
- PhD in a relevant STEM field (e.g., Computer Science, Electrical/Mechanical Engineering, Robotics, or related discipline), or a Master’s degree with equivalent industry experience in applied robotics or robot learning.
- Ability to translate ambiguous embodied AI problems into well-scoped experiments, and maintain rigorous evaluation, ablation, and statistical validation practices.
- Track record of production-ready robotics systems, open-source contributions, or publications in top-tier robotics/AI venues.
- Must be able to report to the MTV office three times per week or any other frequency dictated by the business.
- Must be eligible for relocation benefits.
