Robotics Software Engineer III, Manipulation

About this role

Agility’s commercially deployed humanoids operate alongside teams in warehouses, manufacturing facilities, and distribution centers—tackling physically demanding and repetitive tasks while enabling workers to focus on higher-value work. With industry-leading safety standards and years of proven deployment data, we're pioneering a new era of automation that enhances human potential.

About the Role

We are looking for a Robotics Software Engineer III to join our Skills team and develop manipulation behaviors for humanoid robots. You will own a task vertical from definition through reliable on-robot execution, working across learned and classical manipulation approaches.

You will run our LfD flywheel end-to-end for manipulation tasks: collecting data, defining operator instructions, improving data quality, setting up grading rubrics, training policies, evaluating failures, and iterating until the behavior is robust enough for real deployments.

This role is well suited for an engineer who enjoys working close to the robot and being methodical about failure analysis. You will replay logs, identify weak points in the state distribution, root-cause failures, and build the engineering around policies and primitives so the robot can execute complete workflows.

Key Responsibilities

• Own a task vertical from definition through reliable on-robot execution

• Work across learned and classical manipulation methods

• Run iterative policy improvements across data collection, annotation, training, evaluation, and deployment

• Work closely with teleoperators to improve data quality and operator instructions

• Develop grading rubrics and evaluation workflows that make progress measurable, where success gets measured by policies that act more reliably on real robots in the real world.

• Root cause failures using logs, replays, and analysis of state distributions

• Integrate learned policies with higher-level skills so they can execute inside complete robot workflows

• Build general manipulation capabilities that reduce the amount of task-specific tuning

• Experiment with off-the-shelf manipulation and perception models

• Partner with the ML team that owns the base large behavior model, LfD flywheel, and training recipes used to post-train policies for specific manipulation tasks

About You

• MS or PhD in Robotics, AI, Computer Science, Machine Learning, or a related field

• 2+ years of experience deploying manipulation behaviors on real robots

• Strong software engineering fundamentals and proficiency in Python

• Experience with robot data collection, training, and testing on hardware to perform manipulation tasks

• Familiarity with at least some of the following: grasp planning, imitation learning, reinforcement learning, force control, teleoperation, robot sensors, visual servoing, and object perception

• Strong analytical, experimental, and debugging skills

• Comfortable working hands-on with robotic systems in a lab environment

<span style="font-size: 10pt