Internship | Learning to Chase Objects with a Drone VLA

TNO; Den Haag; On-site; EUR 7,300 - 11,000; Full time

Job summary

TNO is offering a Master’s internship focused on teaching drones to autonomously chase moving targets using a vision‑language‑action model. You will work with simulation tools, build a planner that leverages 3D scene knowledge, and train models using only camera inputs.

The role blends software, ML research, and robotics in a hands‑on environment. You’ll gain experience with Isaac Sim or AirSim, develop demonstrations from a privileged planner, and contribute to autonomous drone perception and

Qualifications

• Open to Master’s students in AI/CS/Robotics with strong ML background.
• Solid Python skills and willingness to work hands-on in research.
• Experience with deep learning frameworks, simulation, or RL/IL is a plus.

Responsibilities

• Set up a simulated environment with a drone and moving targets.
• Develop or configure a planner using 3D scene knowledge to generate demonstrations.
• Train and evaluate a vision–language–action model using camera inputs only.
• Code, run simulations, debug, and iterate on demonstrations and training setups.

Skills

Python

Deep learning (PyTorch)

Simulation environments

Reinforcement learning

Imitation learning

Computer vision

Education

Master's degree in AI/CS/Robotics

Tools

PyTorch

Isaac Sim

AirSim

Job description

Internship | Learning to Chase Objects with a Drone VLA

Make your mark on our time. Become an intern at TNO!

At TNO, the Netherlands Organisation for Applied Scientific Research, researchers work on turning cutting-edge science into technology that matters in the real world — from healthcare and energy to defense and mobility. Within TNO's work on autonomous systems, one recurring challenge is teaching machines to see, understand, and act in complex, unpredictable environments — the same kind of everyday messiness that makes tasks trivial for humans but remarkably hard for robots. This internship sits within that effort, focusing on drones that can autonomously track a moving target — like a car driving down a winding, tree-lined road — using only what they see through their camera.

You will study a vision-language-action (VLA) model that learns to fly a simulated drone in pursuit of a car driving along a winding, tree-lined road, trained entirely on demonstrations generated by a privileged trajectory planner rather than human pilots. Because the planner has full access to the simulator's 3D geometry — the road's curvature, the position and canopy of every tree, the car's future path — it can compute near-optimal chase trajectories that continuously trade off three goals: keeping the target maximally visible (unoccluded, well-framed, at a good viewing angle), holding a minimum safe standoff distance, and steering clear of trunks and overhanging branches. These optimized trajectories, paired with the drone's onboard camera views, become the action-labeled demonstrations used to train the VLA through imitation learning, so the model must learn to infer, from pixels alone, the same 3D-aware flying behavior the planner computes from ground truth — swinging wide before a bend, dipping under a canopy gap, easing off when foliage thickens ahead. The target's behavior stays deliberately simple (a scripted or gently randomized drive along a curved route), and the obstacle layout stays simple in kind but dense in placement, so that the difficulty lives entirely in the occlusion-and-standoff trade-off rather than in modeling a clever adversary — keeping the project's core question sharply focused: can a VLA distill an expert planner's privileged 3D reasoning into a pure vision-and-language-driven flying policy?

What will be your role?

You'll get hands‑on experience with simulation, robotics, and modern AI models, working alongside researchers who build and study these systems every day — a great way to get a feel for both cutting‑edge AI research and what it's like to work at a research organization like TNO.

In practice, you'll set up a simulated environment (e.g., Isaac Sim or AirSim) with a drone, a car driving along a curved, tree‑lined road, and enough obstacles to make chasing genuinely tricky. You'll build or configure a planner that uses full knowledge of the 3D scene to compute optimal chase trajectories — balancing visibility, safe distance, and obstacle avoidance — and use these as training demonstrations for a vision‑language‑action model, which you'll train and evaluate using only camera images as input, much like a real drone would perceive the world. Day to day, this means a mix of coding, running simulation experiments, debugging model behavior, and iterating on what \"good\" demonstrations and training setups look like — the exact methods and scope can evolve as you get into the work and discover what's interesting or tractable. This contributes directly to TNO's broader research on autonomous drone perception and control, and along the way you'll build hands‑on experience with simulation tools, modern AI architectures, and reinforcement/imitation learning methods, while also learning what it's like to shape and scope a research problem as it unfolds — a skill just as valuable as any specific technique.

What we expect from you

This internship suits a student in a Master's programme like Artificial Intelligence, Computer Science, Robotics, or a related technical field, with a solid foundation in machine learning and a genuine interest in computer vision and autonomous systems. We're looking for someone ambitious and self‑driven — comfortable diving into a research problem that isn't fully solved yet, and motivated by the idea of pushing the work toward a publishable result. Strong Python skills are important, and prior experience with deep learning frameworks (e.g., PyTorch), simulation environments, or reinforcement/imitation learning is a real plus, though not something you need to have fully mastered from day one. What matters most is curiosity, persistence, and a hands‑on, experimental mindset — someone who enjoys iterating, debugging, and figuring things out rather than expecting a fixed recipe. You'll fit well in our team if you like thinking independently while also engaging in regular discussion and feedback with your supervisors. The internship typically runs 6-12 months, in line with a Master's thesis, though the exact timing and structure can be discussed. Throughout, there's plenty of room to grow: you'll deepen your skills in modern AI architectures, simulation‑based robotics research, and scientific writing, and if the work reaches a strong enough result, we'll actively support you in shaping it into a publication.

We ask you to include the following information in your application:

• Whether you are looking for a graduation project.


• Your preferred duration (6-12 months).


• Your preferred start date (please note that a security screening may affect the earliest possible start date).

What you'll get in return

An internship at TNO means working in an environment where substance and impact are central. You will become part of a knowledge organisation where research and practice come together, and where experts collaborate on solutions to current societal and technological challenges.

Your internship is a period in which you can discover what suits you, where your strengths lie and what you would like to learn next. You are part of a professional working environment, gain insight into how things work in practice, and have the opportunity to build experience that goes beyond this internship alone. For many students, an internship is therefore also a first step in discovering whether TNO could be a potential next step after graduation.

In addition, we offer you:

• A professional and innovative internship environment in which you actively contribute to societal and technological challenges, working alongside leading experts in your field.


• Personal and dedicated supervision, with focus on