Distinguished Engineer, Trajectory Generation

General Motors•Sunnyvale, CA

About The Position

Overview: We are building a nationwide L3/L4-capable driverless product that is safe, comfortable, and robust across diverse U.S. highway conditions — all under strict real‑time and onboard‑compute constraints. As Distinguished Robotics Engineer for Trajectory Generation, you will own the technical vision, architecture, and quality bars for the planning stack from sensors to trajectories including but not limited to a modern ML planner with multi-stage training. You will design and ship real‑time planners that meet tight latency and compute budgets while delivering best‑in‑class safety, comfort, and robustness, and you will set the experimentation standards and roadmap for this critical part of the autonomy system.

Requirements

10+ years of experience in trajectory generation, motion planning, or related robotics/embodied AI domains.
MS or PhD in Machine Learning, Robotics, Computer Science, or a closely related field.
Deep expertise in trajectory generation for embodied AI, including:
Real‑time planning under kinematic and dynamic constraints.
Integration with perception, localization, and control.
Proven results reducing long-tail errors via:
Targeted data curation and hard-case mining.
Architecture and training changes that move downstream driving KPIs.
Strong background in large‑scale imitation learning (off-policy and on-policy) using fleet-scale data.
Solid understanding of real-time and onboard deployment constraints and how model design impacts latency, memory, and reliability.
Experience with uncertainty estimation, distribution shift, and failure analysis, with strong judgment on model vs. data interventions.
Deep understanding of machine learning foundations and latest AI trends, especially in planning, generative modeling, and RL.
Proficiency with modern deep learning frameworks and distributed training (e.g., PyTorch/JAX, large-scale experiment infrastructure).
Excellent communication skills with both technical and non-technical stakeholders.

Nice To Haves

Experience with foundation models (e.g., VLAs) and their integration into planning or decision‑making stacks.
Prior work on production‑grade autonomous driving, robotics, or other safety‑critical systems.
Demonstrated ability to influence cross‑functional roadmaps and drive alignment across research, product, and platform teams.
Strong background with generative model methods (e.g., Flow Matching) as applied to trajectory proposals or behavior generation.
Strong background in reinforcement learning (e.g., GRPO or similar methods) for policy optimization and rare‑event handling.

Responsibilities

Planning Architecture & Ownership Lead the ML trajectory generation planning stack for nationwide L3, including technical vision, architecture, interfaces, and quality bars.
Define clear APIs and contracts with perception, localization, mapping, and control teams to ensure reliable end‑to-end behavior.
Balance safety, comfort, robustness, and compute constraints in all planning design choices.
Real-Time Trajectory Planning & Safety Design and deploy real‑time trajectory planners that meet on‑vehicle latency, compute, and availability budgets while operating nationwide.
Ensure planning behavior achieves and sustains target KPIs, including intervention rate (MPI), comfort metrics, and reduction in near‑miss collisions (NMC).
Integrate confidence and distribution‑shift indicators directly into the planning loop so low‑confidence or out‑of‑distribution conditions are surfaced and handled appropriately (e.g., safe fallback, policy adjustments, escalation).
Guided Generative Trajectory Proposals & RL Advance planning performance using guided generative trajectory proposal methods to reduce compounding errors and improve rare‑event handling.
Apply reinforcement learning (e.g., GRPO) and related techniques to optimize long‑horizon behavior and sensitivity to nuanced objectives (safety, comfort, efficiency).
Define and run large‑scale experiments combining imitation learning, generative models, and RL over fleet data.
Data, Evaluation & Long-Tail Generalization Leverage fleet‑scale data and hard‑case mining to close long-tail gaps tied to KPI lift (e.g., MPI, NMC) and ensure robust generalization nationwide.
Stand up and maintain a reproducible open‑loop evaluation suite (scenarios, metrics, dashboards) to enable pre‑merge checks and weekly KPI reviews.
Drive a culture of rigorous ablations and failure analysis, with clear guidance on model vs. data fixes and targeted interventions for high‑impact failure modes.
Technical Leadership & Roadmap Influence Act as the technical owner for trajectory generation, setting strategy, experimentation standards, and long‑term roadmap.
Influence the autonomy roadmap and technical trade‑offs, ensuring planning direction is aligned with product goals, safety requirements, and operational constraints.
Mentor senior and principal engineers, raising the bar on planning, trajectory generation, and embodied‑AI practices across the organization.
Communicate complex technical topics clearly to technical and non‑technical stakeholders