Research Scientist - AI Enabled Decision-Making

Job Description

Help define how intelligent systems reason, plan, and act under uncertainty — work that sits at the core of how our mission systems counter swarm threats.

\n

About Praetorian

\n

Praetorian Aeronautics builds advanced autonomous aerial systems designed to protect and defend against the rapidly evolving threat of drone warfare. Headquartered in Adelaide, with offices in Melbourne and Darwin, we develop an integrated ecosystem of counter-autonomy systems spanning high-speed interceptors for kinetic neutralisation of drone threats at range, and AI-enhanced command and control systems that let operators deploy interceptors at scale. Together, these systems enable defence operators to detect, assess, and defeat autonomous threats while maintaining situational dominance in contested environments.

\n

Why This Role Matters

\n

Before an autonomous interceptor is even launched the context of this situation must be understood. Which system pursue which threat, how do interceptors coordinate with other assets, often with incomplete information and under tight resource and communication constraints. Getting this right is a genuinely hard decision-making problem, and decisions need to be explained to a human operator. This role exists to address these challenges: bringing rigorous, research-grade thinking on AI for decision making, whether that be MDPs, combinatorial optimisation, foundation models or multi-agent coordination (to name a few), into systems that are deployed.

\n

The Role

\n

You'll join as a research scientist working closely with our flight sciences and software engineering teams to design and validate decision-making algorithms for autonomous and semi-autonomous operation. The work spans formulating problems as MDPs or POMDPs, applying and extending techniques like reinforcement learning, Monte Carlo tree search, and value or policy iteration, and pairing those with combinatorial optimisation methods for resourcing and task-allocation problems with large, structured action spaces.

\n

You'll move between research and implementation, prototyping in Python, validating against real operational scenarios, and working with the wider engineering team to get promising approaches into deployable form.

\n

We're especially interested in people whose prior work sits at the intersection of novel algorithms and novel applications: publications at venues like IROS, ICRA, AAMAS, or ICML; sequential decision-making over combinatorial action spaces; or distributed decision-making under communication constraints are all directly relevant to the problems we're solving.

\n

What You'll Do

\n

\n
• Formulate and solve sequential decision-making problems as MDPs or POMDPs using RL, MCTS, and value/policy iteration
\n
• Design combinatorial optimisation approaches (LP/IP, genetic algorithms, greedy heuristics) for single- and multi-objective resourcing and allocation problems
\n
• Prototype and validate algorithms in Python using standard ML/DL tooling (Scikit-Learn, PyTorch, etc.)
\n
• Extend approaches into partially observable and multi-agent settings where relevant
\n
• Work with flight sciences and engineering teams to translate research into deployable decision-making capability
\n
• Contribute to the team's publication record where opportunities align with commercial priorities
\n

\n

What We're Looking For

\n

\n
• PhD in Machine Learning or a closely related field, ideally supplemented with further research experience (post-doc, industry research lab, or an ongoing publication record)
\n
• Demonstrable experience with MDP-based decision-making techniques — reinforcement learning, Monte Carlo tree search, or value/policy iteration
\n
• Experience with single- and multi-objective combinatorial optimisation (e.g. linear/integer programming, genetic algorithms, greedy heuristics)
\n
• Strong Python skills and proficiency with ML/DL libraries such as Scikit-Learn and PyTorch
\n

\n

Nice to Have

\n

\n
• Experience in partially observable or multi-agent decision-making domains
\n
• C++ or Rust
\n
• Experience working with LLMs
\n
• Experience with defence applications
\n
• Experience with path planning algorithms
\n

\n

Security & Compliance Requirements

\n

Due to the nature of our work, candidates must be eligible to work in Australia and hold citizenship of Australia or another Five Eyes nation (Australia, United States, United Kingdom, Canada, New Zealand). Security clearance is not required for this role.