Tech Lead, Robotics Runtime & Middleware

Job description

Job responsibilities

• Design and build the robotics runtime/middleware layer – including topic/message architecture, QoS policies, and multi-process communication under tight latency budgets
• Own the inference-serving path for onboard models (VLA/action-expert policies): batching, quantization, hardware acceleration, and the tradeoffs between model accuracy and control-loop latency
• Architect multi-sensor synchronization and fusion arriving on different clocks and cadences, kept coherent enough for closed-loop control
• Make and own the hard technical calls on middleware choice, process/thread architecture, and how much real-time guarantee any given subsystem actually needs
• Set technical standards and do deep design/code review across runtime engineering
• Prototype and de-risk new architecture directions (new middleware, new compute hardware, new model-serving approaches) before they become team-wide commitments
• Work directly with modeling/research to understand what a policy actually needs from the runtime (latency, synchronization, action representation) and make sure the platform delivers it

Job requirements

• Strong systems programming in Python and/or Rust
• Hands-on experience with robotics middleware — ROS2 (rclcpp/rclpy, DDS implementations like Fast DDS/CycloneDDS, or Zenoh), including designing custom message types, QoS tuning, and multi-node communication patterns
• Real-time systems experience: understanding of scheduling, latency budgets, jitter, and the difference between soft and hard real-time guarantees; comfort with RTOS concepts
• Experience with embedded/edge Linux — cross-compilation, device driver basics, and resource-constrained execution (CPU/memory/power budgets)
• Multi-sensor synchronization: time sync protocols (PTP/gPTP), sensor fusion pipelines, and handling clock drift/skew across distributed compute on a single robot
• Model deployment/inference optimization: exporting and running models via ONNX Runtime, TensorRT, or similar; quantization and batching strategies for action-model inference on edge compute;
• Experience with control systems and motion primitives — enough to reason about how software architecture decisions (message latency, action chunking, control frequency) affect closed-loop robot behavior
• Track record of taking real-time systems from prototype to reliable, continuous operation on physical hardware