At NEURA Robotics, cognitive robots don’t just execute commands. They perceive their surroundings, understand them, reason, and act autonomously in the physical world. Our robotic foundation models handle the physical execution. Above them sits the embodied agent: the part that interprets what the robot sees, maintains an up-to-date world state, and uses that understanding to decide what to do next. We’re looking for an AI Engineer with strong hands-on experience in visual perception and reasoning to build this environmental awareness for our humanoid platform and connect it to reliable task planning and execution.
Your mission & challenges
Visual reasoning & world state: You build the robot’s understanding of its surroundings using visual perception and VLMs. You ground language in the scene, reason about spatial relationships and affordances, and maintain a working representation of objects, their states, and the task history as new observations arrive. You make uncertainty and environmental changes available to the agent, so its decisions stay grounded in what is actually happening.
Task decomposition & long-horizon planning: You turn high-level human intent into ordered, executable steps, each with the preconditions, spatial references, and success criteria it needs. You sequence them so the robot can work through a task it has never seen in that exact form before.
Closed-loop execution & recovery: You verify each step against observation instead of assuming success. You detect divergence from the plan and replan, retry, or escalate to a human when the risk is too high.
Evaluation on the robot: You evaluate scene understanding, visual grounding, world-state consistency, and task success in simulation and on real hardware. You build failure taxonomies from real runs and use them to distinguish perception and reasoning failures from planning, interface, and execution failures.
Embodied agent harness: You design and build the agent that runs on our humanoid platform: the loop from sensor input to reasoning to action, its interfaces to cameras, depth sensing, and proprioception, and the state management that keeps a task coherent over minutes rather than seconds.
Cross-functional collaboration: You work closely with our foundation model, perception, manipulation, and hardware teams to take the reasoning layer from concept to reliable behavior on a deployed robot.
What we can look forward to
Education: A Master’s or PhD in Computer Science, Robotics, or a related field, or equivalent depth demonstrated through shipped work.
Visual perception & VLMs: Strong hands-on experience with visual perception and VLMs: grounding language in images, understanding objects and their states, spatial and physical reasoning, and producing structured scene representations an agent can act on. You have integrated these capabilities into working systems and understand their failure modes under ambiguity, occlusion, and changing viewpoints.
Agent systems: Demonstrated experience building agent systems on top of LLMs and VLMs that decompose tasks, plan, use tools, and maintain state over long horizons, rather than single-turn prompt work.
Work on physical robots: Several years of hands-on work on physical robots, not only in simulation. You bring your own code onto a real platform, run it there, and debug a system that behaves differently once it has a body.
Hierarchical autonomy: Familiarity with hierarchical autonomy, where a reasoning model plans and learned or classical policies execute: subgoal interfaces, progress verification, and replanning.
Middleware & onboard deployment: Experience with robotics middleware and onboard deployment: ROS 2, inference on embedded GPUs, and working within a real latency budget.
Software engineering: Solid engineering practice: clean code, testing, and services other teams can build against. Cloud, containerization, and CI/CD experience is a plus.
Languages: Excellent English. German is a plus.