Principal Robotics Learning Engineer

AIM•Seattle, WA

About The Position

AIM builds autonomy for the real world—robots that move mountains. Our systems bring together software, hardware, and robotics to turn heavy equipment into intelligent machines that reshape how earthmoving is done. Operating these machines requires skills that adapt to changing terrain, material properties, loads, and contact conditions. We’re looking for a Staff-level or more senior Robotics Learning Engineer to lead the technical development of data-driven autonomy skills, helping our machines learn and improve from demonstrations, simulation, and experience. This is a hands-on individual-contributor role on our technical career track, with level matched to experience and scope at Staff, Senior Staff, Principal, or above. We will build learned autonomy skills for heavy equipment, with potential applications such as digging, loading, grading, and interacting with variable materials. We will connect these skills to the broader autonomy system and use machine data to improve their reliability over time. Our measure of progress will be repeatable task performance: successful execution, useful productivity, fewer interventions, and reliable operation across changing conditions.

Requirements

Demonstrated technical ownership at Staff level or above of substantial robotics or robot-learning efforts: setting direction, making architecture decisions, coordinating across teams, and delivering working systems.
A track record of personally implementing, training, deploying, and evaluating learned policies on physical robots or vehicles. Depth in reinforcement learning, imitation learning, or behavior cloning is required; experience across RL and BC is welcome.
Recent hands-on contributions to production or deployment-quality software, learning experiments, and hardware debugging, alongside technical leadership.
Experience mentoring engineers and improving the quality of technical decisions through design and code review.
Strong understanding of machine learning, optimization, and sequential decision-making, including generalization and distribution shift.
Proficiency in Python and a modern learning framework such as PyTorch or JAX.
Experience building reproducible data, training, and evaluation workflows and interpreting experimental results.
Working knowledge of robotics sufficient to reason about observations, actions, dynamics, feedback, and physical constraints.
Ability to write maintainable software, diagnose failures across system boundaries, and communicate technical tradeoffs clearly.
A degree in Robotics, Computer Science, Electrical or Mechanical Engineering, or a related field, or equivalent practical experience.

Nice To Haves

Sustained field or customer deployment of learned policies, including performance monitoring, regression prevention, and improving reliability after launch.
Experience with demonstration collection, teleoperation, dataset design, offline RL, policy fine-tuning, or iterative data collection.
Experience with simulation-to-reality transfer, system identification, domain randomization, or learning for contact-rich tasks.
Experience with visuomotor policies, diffusion policies, action-sequence models, or hierarchical and multitask skill learning.
C++ and experience integrating learned models into a robot software stack with real-time constraints.
Experience applying robot learning to off-road machines, humanoids, manipulation, locomotion, or other systems operating under variable loads and contact conditions, with the ability to transfer those methods to heavy equipment.
Hydraulic or electrohydraulic systems experience is a plus, not a requirement.

Responsibilities

Lead the technical effort
Build and evaluate a proof of concept (POC) for a representative autonomy skill to establish feasibility and a measurable performance baseline.
Use the POC to identify gaps in data, simulation, learning methods, machine integration, and evaluation, and prioritize what must improve.
Develop and execute a phased technical roadmap based on the POC findings, from closing critical gaps through deployment and continuous improvement.
Define the architecture and interfaces across data, simulation, training, policy inference, controls, and evaluation.
Make and communicate tradeoffs between RL, imitation learning, behavior cloning, and established robotics methods based on task needs, data availability, and measured results.
Lead cross-team execution, resolve difficult technical blockers, mentor engineers, and personally implement critical parts of the system.
Develop learned autonomy skills
Formulate robot-learning problems: observations, actions, task goals, rewards or imitation objectives, and policy structure.
Implement and evaluate reinforcement learning, behavior cloning, or related imitation-learning approaches; select and combine methods when the evidence supports it.
Build skills that handle variation, recover from failures, and work with existing planning and control systems.
Build the data and training loop
Work with operators and robotics engineers to collect demonstrations, machine trajectories, interventions, and failure cases.
Develop tools for synchronization, dataset curation, quality checks, versioning, and reproducible training.
Identify gaps in the data and design targeted collection or simulation experiments to improve coverage and robustness.
Transfer policies to real machines
Build and improve simulation-based training and testing where useful, and characterize the differences between simulated and physical behavior.
Address deployment constraints such as sensor noise, latency, action limits, inference performance, and changing machine dynamics.
Integrate learned policies with perception, state estimation, controls, and runtime supervision, including defined fallbacks and transitions between operating modes.
Establish reliable evaluation
Compare learned policies with appropriate baselines using held-out tasks, environments, or operating conditions.
Measure task success, intervention rate, cycle time, constraint violations, and other relevant machine outcomes.
Diagnose failures from logs and experiments, and build regression tests that catch performance losses across model and data updates.
Validate changes through simulation and controlled hardware trials before wider field deployment, working with systems and safety engineers.