Test Engineer — Dexterous Humanoid Hand (human)

Your mission & challenges

We are building dexterous humanoid hand platforms — multi-DOF tendon-driven systems with embedded tactile sensing, real-time motor control, and AI-driven perception running on edge compute. The stack spans custom PCBs, bare-metal firmware, ROS 2 middleware, and learned inference models, all converging in a physical system that must work reliably in the real world.
We are looking for a Test Engineer who can own validation across that full stack. Not a QA function that arrives at the end of a development cycle — someone who builds the test infrastructure from the ground up, defines what "working" means at every layer, and keeps the team honest throughout development.
The programme is early-stage. There is no existing test framework to inherit. You will design it.

• Hardware-Software Integration Testing

  • Design and execute integration test protocols for the hand control stack: motor driver boards, embedded compute modules, sensor interfaces, and the communication backbone (DDS over TSN Ethernet and SPI)

  • Develop hardware-in-the-loop (HIL) test rigs for validating firmware behaviour against real actuator and sensor hardware — brushed DC and BLDC motor channels, absolute encoders, tactile sensor arrays, IMUs

  • Define and automate bring-up test sequences for new PCB revisions: power-on checks, bus enumeration, driver smoke tests, and channel-by-channel functional validation

  • Own the integration test protocol for the forearm-to-body elbow interface: DDS topic correctness, latency measurement, link-loss behaviour, and safe-state transitions under fault injection

  • Test the full closed-loop control pipeline end-to-end: sensor input → embedded inference → motor command → physical response, with instrumented ground truth at each stage

  • Instrument and measure system timing: control loop jitter, DDS publish latency per topic, NPU inference latency, and end-to-end perception-to-action latency against defined SLAs

  • Validate mechanical-electrical interfaces: connector continuity through range of motion, cable harness stress testing, signal integrity under flexion cycles

• Software Testing

  • Build and maintain the SW test suite covering: ROS 2 nodes and DDS topic pipelines, motion primitive state machines, grasp sequencer logic, and safety watchdog behaviour

  • Design unit and integration tests for the embedded inference pipeline: ONNX model output correctness versus CPU reference, ring buffer behaviour, multi-task DDS publishing under sustained load

  • Implement regression test coverage for the control stack: position control loop stability, force ceiling enforcement, corrective tighten response timing, and arbitration logic between concurrent control modes

  • Define and run fault injection tests in software: simulate link loss, sensor dropout, classifier confidence below threshold, consecutive high-severity slip events — confirm the system transitions to the correct safe state in every case

  • Build simulation-based tests where physical rigs are unavailable: URDF-based motion validation, kinematic limit checking, and trajectory feasibility before hardware deployment

  • Maintain CI pipeline integration: automated test runs on every firmware and software commit, with clear pass/fail gates and failure triage

  • Own the benchmark test protocol for external hand platforms: define repeatable, instrumented test procedures

• Test Infrastructure

  • Select and deploy appropriate test tooling across the stack: logic analysers, oscilloscopes, force/torque sensors, motion capture or camera-based ground truth, data loggers

  • Build a structured test results database: every test run logged with software version, hardware revision, configuration, and outcome — traceable and queryable

  • Write test specifications that other engineers can execute independently and reproduce your results

  • Define acceptance criteria for each subsystem before integration begins — not after

What we can look forward to

• Essential

  • Degree in electrical engineering, mechatronics, computer science, or a related field
    Hands-on experience testing embedded systems: bring-up, bus protocols (SPI, I2C, UART, CAN), signal integrity measurement, and firmware debugging with real hardware

  • Experience writing automated software tests in Python or C++ — unit tests, integration tests, regression suites

  • Ability to read and understand firmware and software well enough to identify what needs testing and where the edge cases are, without needing to write all the production code yourself

  • Experience with instrumentation and measurement: oscilloscopes, logic analysers, current probes — comfortable setting up a bench and capturing what is actually happening

  • Structured thinking about failure modes: given a system description, you should be able to enumerate the ways it can fail and design tests that would catch them

• Strongly Preferred

  • Experience with ROS 2, DDS middleware, or real-time communication systems

  • Experience with hardware-in-the-loop testing or physical test rig design

  • Familiarity with motor control systems — brushed DC or BLDC — and the characteristic failure modes of position control, current limiting, and encoder feedback loops

  • Experience building or contributing to CI/CD pipelines for embedded or robotics software

  • Familiarity with force/torque measurement and tactile sensing systems

  • Experience testing learned or probabilistic systems: validating model output distributions, testing confidence thresholds, and defining acceptable behaviour under out-of-distribution inputs

  • Background in or exposure to functional safety testing — understanding the difference between testing for correctness and testing for safety is a significant advantage

• What Makes This Role Unusual

  • Most test roles in robotics are either purely software or purely hardware. This one spans both and demands genuine fluency at the boundary — the interesting failures in a system like this are almost always at the interface between a PCB revision, a firmware change, and a DDS schema update happening simultaneously.

  • You will also be testing a system where the perception layer is a learned model running on an NPU. Defining what "correct" means for a tactile slip classifier feeding a safety-relevant motor command is not a solved problem. You will need to think carefully about how to validate probabilistic outputs in a deterministic control context, and how to define test coverage for a system that can fail gracefully rather than fail cleanly.

  • The benchmark framework for external hand platforms is also yours to own. That work has direct strategic value — the results inform which external platforms we integrate and how our own platforms compare. It is high-visibility work with real programme impact, not a back-office function.