Autonomy Algorithm Engineer

Develop the core state estimation, navigation, control, and mission-level autonomy algorithms

enabling fully autonomous UAV flight. Covers SLAM, Visual-Inertial Odometry (VIO), Kalman

filter design, GNSS-denied navigation, and mission-level behavioral autonomy for operations in

contested, GPS-degraded, and communications-denied environments.

Reports to: Head od Software & AI

‍

State Estimation & GPS-Denied Navigation:

● Design Kalman filter (extended and unscented) and factor-graph estimators fusing

inertial, GPS, vision, barometer and radio-based positioning into a robust six-

degree-of-freedom pose estimate.

● Build and tune visual-inertial odometry and visual SLAM (simultaneous localisation

and mapping) pipelines - ORB-SLAM3, VINS-Fusion, LIO-SAM or custom - for

localisation in GPS-denied and contested environments.

● Ensure the estimator degrades gracefully and recovers from sensor dropout, GPS

jamming and spoofing, including data-driven detection of these conditions to trigger

fallback modes.

● Where it measurably beats the classical baseline, introduce learned components

into the estimator: adaptive noise models, inertial error correction or neural feature

matching.

Flight Dynamics & Trajectory Algorithms:

● Develop motion prediction, trajectory estimation and adaptive filtering for dynamic

flight, sensor degradation and environmental uncertainty.

● Create estimators supporting target tracking, collision avoidance and cooperative

swarm behaviour, interfacing with the flight controller over MAVLink/PX4.

● Apply learned or hybrid dynamics models where physics-based models under-

perform: aggressive manoeuvring, aerodynamic interaction or degraded actuation.

Mission-Level Autonomy:

● Design mission behaviours — persistent surveillance loiters, autonomous threat-

response replanning, return-to-safe-area under sensor failure or loss of

communications - and adapt them in flight as sensors degrade, threats emerge or

target assignments change.

● Apply reinforcement learning to mission decisions taken on incomplete information

(adaptive loiter, sensor tasking, threat response), feeding the estimator's own

uncertainty into the decision rather than assuming perfect knowledge of the world.

● Train those policies in simulation, transfer them to hardware, and measure them

honestly against conventional rule-based planners before they fly.

● Bound every learned behaviour with hard safety limits and a deterministic fallback,

so aircraft behaviour stays predictable and certifiable.

● Develop terminal guidance algorithms for precision terminal-phase operations,

including proportional and augmented navigation under high-dynamic flight.

● Integrate geofencing, airspace deconfliction and dynamic no-fly zone enforcement

to EASA U-Space requirements.

Integration & Validation

● Integrate estimation modules with the flight stack (PX4/ArduPilot) and ROS 2

autonomy pipelines.

● Validate in simulation (Gazebo, MATLAB/Simulink, PX4 software-in-the-loop) and

on hardware-in-the-loop rigs before field trials.

● Deploy any neural inference to onboard computers (NVIDIA Jetson class) within

real-time, power and thermal budgets.

● Analyse telemetry against measurable targets - localisation drift, convergence

time, robustness to sensor failure - and iterate.

● Document designs, derivations and interfaces to support certification.

‍

‍

‍Required:

● Master's or PhD in Robotics, Control Systems, Electrical Engineering, Computer

Science or a related field.

● 3+ years developing state estimation or autonomous navigation for drones,

robotics or autonomous vehicles.

● Proven Kalman filter design (extended and unscented), sensor fusion architecture

and probabilistic motion modelling.

● Deep knowledge of SLAM - visual, LiDAR or fused - and visual odometry.

● C++ and C (Eigen, Ceres, GTSAM) and Python (NumPy, SciPy).

● Working knowledge of machine learning and practical use of PyTorch or similar:

building datasets, training, validating, and judging whether a model genuinely

generalises.

● Understanding of reinforcement learning basics - framing a decision problem,

designing a reward, choosing between value- and policy-based methods - plus

the judgement to recognise when a conventional algorithm is the better answer.

● Simulation and hardware-in-the-loop testing (MATLAB/Simulink, Gazebo, PX4).

● Familiarity with PX4/ArduPilot flight stacks and the MAVLink protocol.

● Strong mathematics: linear algebra, probability, numerical methods, optimisation,

control theory.

● Version control discipline (Git/GitLab).

● English at upper-intermediate level or above, and the ability to work with adjacent

engineering teams.

● National of a NATO member state, or of Australia, Japan, South Korea, New

Zealand or Ukraine.

● Clean criminal record.

Preferred:

● ROS 2: package development, transform trees, sensor integration, data pipelines.

● Hands-on reinforcement learning for robotics - training policies with standard

algorithms and frameworks (for example PPO or SAC in Gymnasium, Stable-

Baselines3 or Isaac Lab) — or imitation learning to bootstrap policies from

demonstrations.

● Sim-to-real transfer: domain randomisation, system identification, curriculum

learning.

● Swarm work of any kind: multi-agent localisation, distributed optimisation, or multi-

agent reinforcement learning.

● Learning-augmented estimation: differentiable filters, learned inertial odometry, or

neural SLAM front-ends.

● Running neural inference efficiently on embedded GPUs (Jetson Orin, TensorRT,

quantisation).

● Tightly-coupled visual-inertial odometry (VINS-Mono, Kimera, OpenVINS) and

factor graph optimisation (GTSAM, g2o).

● Radio-based positioning (ultra-wideband, pseudolite) as a supplementary

navigation source.

● Real-time operating systems (FreeRTOS, Zephyr, NuttX) for hard real-time

estimation loops.

● Safety-critical standards: DO-178C, MISRA C/C++, STANAG 4671, and EASA

guidance on machine learning in aviation.

● Publications at ICRA, IROS, CoRL or IEEE Transactions on Robotics; open-source

contributions to SLAM, estimation or reinforcement learning projects.

‍