Software Engineer, Robot Perception & Vehicle Integration

Software Engineer, Robot Perception & Vehicle Integration

Tesla

Grünheide (Mark) or Brandenburg, Germany

What to Expect

This team at Tesla Manufacturing Brandenburg in Grünheide builds the autonomy that lets a reach truck operate unmanned across our warehouse operations. We own the full on-vehicle stack: perception and vision-based localization, state estimation and mapping, motion planning and the behaviour layer that sequences navigation and load handling, the real-time interfaces to the vehicle's motion and safety systems, low-latency teleoperation for supervised recovery, and integration with the WMS and FMS that dispatch and track the work. Our remit runs from that software and systems architecture through to a machine that is functionally safe, CE-certified, and proven from a first pilot cell to dependable, everyday operation. The robot has to perceive and pick pallets it has never seen before, without per-SKU models, and it has to stay operable when autonomy hands off to a person. This role owns the perception that makes the first of those possible, along with the software that keeps the vehicle connected to everything around it. Perception is the heart of the job rather than one half of a role split down the middle.

Role mission

You build the perception the robot depends on in the field, the low-latency teleoperation client for supervised recovery, and the vehicle-to-fleet agent that connects it to the warehouse systems. You carry each of those from a first design through to something that works dependably on a real vehicle.

What You’ll Do

• Above all, build the perception that lets the robot handle pallets it has never seen before. From stereo depth and detections you estimate pallet pose and the geometry that matters for a safe fork insertion, including the pick face, the support surface, the fork pockets, and any overhang, favouring conservative, geometry-driven policies over per-SKU models, and you read fiducials to reconcile against what the warehouse system expects. Most of your time goes here, and you own the calibration, the latency budgets, the offline evaluation, and the quality of the data behind it all.
• Build the low-latency teleoperation client (drive, mast, and video) that lets an operator supervise or recover the vehicle.
• Build the vehicle agent that exchanges jobs, state, and telemetry with the existing WMS and FMS, as an adapter rather than a rewrite. This work is real and it matters, but it sits behind perception in priority, and we will not let it crowd perception out.

What You’ll Bring

• 3+ years writing software, and genuinely both sides of this role. You have shipped perception or computer-vision work on a real robot or vehicle, or a strong depth-camera project that had to hold up against real-world sensor noise, lighting, and clutter, and you have also built solid integrations against real APIs and systems.
• Strong C, C++, and Python.
• Comfortable building and running software on Linux, including real-time Linux.
• A solid grounding in computer vision and 3D/depth perception, in camera and hand-eye calibration, and in working with stereo or RGB-D sensors.
• Familiarity with ROS 2, which is the framework we use.
• Strong English, the language the team works in.

Nice to have

• Experience with the kind of stereo-depth cameras used on mobile robots.
• Teleoperation or low-latency video streaming, for example with WebRTC.
• Having integrated with warehouse or fleet-management systems before.
• A comfortable, AI-native way of working, where modern AI coding assistants and evaluations are already part of how you build.