Research Scientist, Architectures

Our AI Research team is building end-to-end robot policies that enable dexterous manipulation in realworld environments. We are advancing embodied AI by integrating multimodal perception, robot learning architectures, and physical execution systems to solve manipulation, autonomy, and simulation challenges at an industrial scale.
As a Principal Applied Scientist, you will define technical direction, create breakthrough robot learning methodologies, and drive the execution of largescale initiatives that deliver endtoend manipulation policies. You will architect new model families, lead crossfunctional research efforts, and guide teams in transitioning scientific advances into reliable robotic capabilities deployed in real systems.
Define the research strategy and technical roadmap for embodied manipulation policies that span perception, training, control, and deployment.
diffusionbased policies, ACTstyle agents, multimodal embodied transformers) to enable robust dexterous manipulation.
Architect endtoend policy systems from multimodal sensing to hardware control, across simulation and real environments.
Establish standards for data generation, demonstrations, simulation fidelity, evaluation, and simtoreal adaptation for manipulation tasks.
Guide crossfunctional efforts to scale robot learning from prototypes to production, including reliability constraints, observability, and performance guarantees.
Mentor scientists and engineers, influence hiring, and shape research priorities within the organization.
Represent GM in external communities (e.g., CoRL, RSS, ICRA, NeurIPS), pursue collaborations, and drive thought leadership in embodied AI.
PhD in a relevant STEM field, with a recognized track record in robotics, robot learning, or embodied AI.
Significant experience designing and deploying robot learning systems on physical hardware, including manipulation tasks and realtime policy execution.
Expertise in modern AI architectures (Transformers, diffusion models, VLM/VLA agents, imitation learning, offline RL) and ability to extend them to new embodied scenarios.
Mastery of PyTorch, including model internals, performance optimization, distributed training, and debugging complex failures.
Handson experience with ROS/ROS2, motion planning stacks, and integration of ML components into robotic platforms.
Evidence of highimpact contributions: toptier publications, influential opensource efforts, field deployments, or foundational algorithms.
Proven ability to lead ambiguous multidisciplinary research programs and deliver tangible embodied AI outcomes.
Experience establishing research agendas in robot learning or dexterous manipulation that resulted in widely adopted systems or protocols.
Leadership of projects involving perception policy control integration, staged training, or multirobot deployment.
Physicsinformed or tactileinformed policies
Foundation models for embodied agents
~ Strong industry or academic presence: invited talks, program committees, major opensource contributions, or collaborations with robotics research institutions.

You will shape how robots behave in the physical worlddesigning the foundational learning systems that enable dexterous, autonomous manipulation across GMs future robotics platforms.
The successful candidate is expected to report to the MTV office three times per week or any other frequency dictated by the business.
Bonus potential: An incentive pay program offers payouts based on company performance, job level, and individual performance.
Benefit options include medical, dental, vision, Health Savings Account, Flexible Spending Accounts, retirement savings plan, sickness and accident benefits, life insurance, paid vacation & holidays, tuition assistance programs, employee assistance program, GM vehicle discounts and more.
Relocation
This job may be eligible for relocation benefits.
#