Senior Agentic Planning Engineer

Lightspeed•Northbrook, IL

10d

About The Position

As Senior Agentic Planning Engineer, you will own how LightSpeed's robots decide what to do next. You will build a planner that reasons over goals, sequences robot actions, chooses between classical control and learned policies, and recovers when actions fail. The planner will make ongoing decisions using live sensor data and the results of prior actions, enabling robots to complete whole panels or rooms autonomously. You will define action primitives and their contracts, build task graphs from BIM, and develop a two-layer planner combining reasoning for open-ended decisions with solvers for reach, collision, and trajectory. You will determine when reasoning models are appropriate, choose the executor for each step, monitor execution, and provide recovery and escalation paths. You will also develop operator tools, prepare the planner for multi-robot work, and prove the system through end-to-end tasks on real robots. You will set planning and autonomy architecture across both robot families, mentor autonomy and motion planning engineers, and remain hands-on by writing production code and testing on real robots every week. The role co-owns foundation interfaces with world-sensing and robot-learning principals and owns the primitive contract. The new stack will run alongside production on existing robots and be switchable per job.

Requirements

5+ years building autonomy, task-planning, or decision-making systems for robots or autonomous vehicles, with systems deployed in production.
Experience with task-level planning on real robots, such as behavior trees, task and motion planning (TAMP), HTN, or similar.
Experience with execution monitoring and fault recovery in a deployed system.
Experience building interfaces between planning, perception, and control that other teams used.
Hands-on experience with reasoning models or LLMs in a decision loop, and a clear understanding of where they are and are not useful.
Strong Python and C++ skills.
Experience with ROS2, Linux, Docker, and CI/CD workflows.
A record of setting technical direction across teams and mentoring senior engineers.

Nice To Haves

MS or PhD in Robotics, Computer Science, AI, or a related field.
Behavior or decision planning experience on an autonomous vehicle team.
Experience with LLM or VLM agents for robotics, including constrained action spaces and guardrails.
Motion planning and trajectory optimization experience with MoveIt, OMPL, or similar.
Experience with multi-robot coordination, task allocation, or fleet orchestration.
Experience with functional safety or formal verification for autonomous systems.
Experience planning from CAD or BIM models.
Background in construction, manufacturing, or industrial automation.

Responsibilities

Define robot action primitives and their contracts, including inputs, preconditions, success checks, and failure codes.
Define the planner-executor interface for step requests, executor selection, results, and evidence.
Maintain a primitive library that the planner can call without task-specific hand wiring.
Build task graphs from BIM elements and their dependencies.
Develop a two-layer planner with a reasoning layer for open-ended decisions and solvers for reach, collision, and trajectory.
Determine when to use reasoning models and keep them off the real-time control path.
Set rules for choosing classical control or a learned policy for each step based on world-model confidence and step structure.
Extend planning and sequencing for new trades and measure success by how quickly a new trade can be taught.
Build an execution monitor that verifies steps and can retry, switch executors, skip and flag, or escalate to a person.
Log each decision with its reason and evidence.
Keep the planner within guardrails, including a constrained action space, declared parameter ranges, and safety controls that are not delegated to the planner.
Build the operator view for escalations and remote assistance.
Design for two robots to share a world model and divide work.
Co-own foundation interfaces with world-sensing and robot-learning principals, and own the primitive contract.
Run the new stack alongside production on existing robots, switchable per job.
Prove the system by running whole tasks end to end on real robots, including recovery.
Set planning and autonomy architecture across both robot families.
Mentor autonomy and motion planning engineers.
Write production code and test on real robots every week.