Reinforcement Learning Engineer

The company

AEGIR is a deeptech start-up developing AI-enabled Command & Control (C2) software to manage complex naval and maritime missions. These missions involve autonomous systems — underwater, surface and aerial drones — in civilian contexts (exploration, inspection, security) and military ones (surveillance, seabed warfare, mine countermeasures).

OCTOPUS, the software developed by AEGIR, supports the design, planning, supervision and conduct of these missions, taking the constraints of both the systems and the environment into account. It comes in three modules:

• Vision, for a Common Operational Picture shared in real time;
• Decision, for coordinating multi-domain systems;
• Action, an embedded AI for autonomous mission execution.

The position

The position is based in Lyon, with 2 days of remote work possible. Occasional travel is expected. You will work on the Decision and Action modules of OCTOPUS, with a focus on coordination and autonomy, alongside the software, algorithm and robotics teams, from model design to validation in simulation and on real platforms.

Responsibilities

• Design and develop Deep Reinforcement Learning algorithms for the autonomy of systems: navigation and decision-making.
• Develop multi-agent reinforcement learning methods for the coordination and cooperation of autonomous systems.
• Develop methods to find optimal trajectories under constraints (energy, time, safety, environment, mission).
• Design mission-planning strategies for one or several autonomous systems (USV, UAV, UUV).
• Develop swarm coordination and cooperation algorithms enabling dynamic task allocation and adaptation to operational contingencies.
• Design simulation environments for the training and evaluation of learning agents.
• Evaluate the performance of learned policies, improve their robustness, generalisation and efficiency, and take safety constraints into account.
• Integrate the developed algorithms into the OCTOPUS products and take part in their validation on real systems.
• Collaborate with the software, robotics and product teams to integrate the solutions into an operational architecture.

Profile

• Higher education degree (Master’s or PhD) in artificial intelligence, robotics, applied mathematics, computer science, control engineering or an equivalent discipline.
• Minimum 3 years of experience in the field.
• Solid skills in Deep Reinforcement Learning, reinforcement learning or sequential optimisation.
• Mastery of Python and of the main Deep Learning frameworks (PyTorch or TensorFlow).
• Knowledge of optimisation, decision theory, operations research or optimal control.
• Experience in, or interest for, trajectory planning, autonomous navigation or multi-agent coordination.
• Ability to work autonomously on applied research topics while collaborating closely with a multidisciplinary team.
• Scientific curiosity, rigour and a taste for technical challenges.
• Professional English.
• Experience with simulation environments (Gymnasium, Isaac Sim, Gazebo, Unity ML-Agents, AirSim, etc.) is a plus.
• Knowledge of methods such as A*, RRT*, MCTS, PPO, SAC, DQN, MADDPG or MAPPO is a plus.

Working environment

• Multidisciplinary team of experts in AI, robotics, optimisation and real-time software.
• Development of breakthrough technologies for next-generation autonomous systems.
• Projects combining applied research, field experimentation and industrialisation.
• Start-up environment fostering autonomy, innovation and fast growth in responsibilities.

Salary and benefits

• Gross annual salary of €45,000 to €55,000, depending on experience and profile.