Autonomy ML Engineer

What this is

SWARM is building the intelligence layer for autonomous drones. We are looking for an Autonomy ML Engineer to build learning-based flying agents that can perceive, decide and act in simulation and, ultimately, on real aircraft.

A core objective is to beat the state of the art on SWARM's autonomous-flight benchmarks, including SOTApilot, and transfer those capabilities into commercial autonomous-drone systems such as autonomous inspection and surveillance.

This is not a generic AI, LLM or prompt-engineering role.

What you will work on

• Train autonomous flight policies for navigation, obstacle avoidance and mission execution.
• Improve benchmark performance across unseen environments, not just known training scenarios.
• Design observations, actions, rewards, curricula and evaluation strategies.
• Run simulation experiments, analyze failures and iterate systematically.
• Improve robustness, safety, speed and generalization of learned policies.
• Develop capabilities for more complex missions: search & rescue, interception and multi-agent coordination.
• Combine learning-based autonomy with classical planning and control where appropriate.
• Help move successful policies from simulation onto real drones.
• Contribute perception and autonomous decision-making capabilities required by commercial projects.
• Documenting SWARM's architecture, inventions, and engineering systems.

What we're looking for

• Approximately 2–4 years, ideally around 3, in a highly relevant professional or research environment. We care more about what you have built than the number of years on your CV.
• Someone who has already trained autonomous or reinforcement-learning systems that actually work, and can independently improve them through structured experimentation.
• An unconventional background with strong demonstrated work counts for more here than a conventional one without it.

Strong requirements

• Strong Python and PyTorch
• Reinforcement learning and continuous control
• PPO, SAC, TD3 or related methods
• Robotics / physics simulation
• Reward design, evaluation and failure analysis
• Structured experimentation and reproducibility
• Robotics fundamentals: position, velocity, orientation, coordinate frames
• Ability to own open-ended technical problems independently

Valuable experience

• Isaac Sim / Isaac Lab, MuJoCo, PyBullet, Gazebo or similar
• Computer vision, depth or learned perception
• Curriculum learning and domain randomization
• Sim-to-real transfer
• Multi-agent reinforcement learning
• Motion planning and control
• ROS2, PX4 / MAVLink
• ONNX / TensorRT / edge deployment
• C++

What matters

• Understand why the current system fails.
• Propose experiments and technical approaches.
• Implement and train them.
• Evaluate results rigorously.
• Iterate until performance measurably improves.

What success looks like

• Higher autonomous-flight benchmark performance.
• Better generalization across unseen environments.
• Fewer collisions, failures and unsafe behaviors.
• Faster and more reliable mission completion.
• Successful progress from simulation toward real flight.
• New autonomous capabilities deployed in customer projects.

Build increasingly capable flying agents.

Compensation

Competitive net compensation based on level, relevant experience and specialist expertise.

Apply

Tell us about the most relevant autonomous, robotics or reinforcement-learning system you've built.