Master Thesis - Benchmarking AI on Resource-Constrained Micro Drones

Micro-drones have tight limits on computing capacity, memory, energy, payload and flight duration, making it difficult to predict whether AI models that work in simulation will perform reliably on real hardware. This master's thesis at Bosch in Lund aims to develop and validate a benchmarking framework for AI deployed on these platforms. The research will identify useful performance metrics and explore trade-offs between model quality and drone constraints through simulation and a bounded hardware test scenario.

The assignment

• Define benchmarking metrics covering model quality, execution performance, resource consumption, energy trade-offs and operational behaviour.
• Review and adapt the existing deployment pipeline to run selected machine-learning inference models on micro-drone hardware.
• Choose available models and establish a controlled scenario and test environment for reproducible comparisons.
• Implement workflows for measurement, logging, analysis and reporting.
• Validate the framework in simulation and on a real micro-drone, comparing relevant configurations and assessing how the metrics inform deployment decisions.
• Deliver a proof-of-concept framework and a documented evaluation method, keeping the scope focused on a specific flight scenario rather than a general autonomous flight solution.

Who we are looking for

• Studying Information Technology, Computer Science, Electronics, Mathematics or Physics.
• Relevant coursework or knowledge in data science, control systems and AI.
• At least basic programming skills in Python, C++ or a comparable language.
• Able to take initiative and advance the project independently.
• Collaborative and communicative, with a genuine interest in robotics, AI and autonomous systems.

What we offer

• Access to existing hardware and software environments, drone components, programming and control interfaces, simulation assets and simulation data.
• Access to knowledge relating to machine learning and AI for drones.
• Opportunity to influence the choice of models, metrics and test scenarios as feasibility and project findings become clearer.

Practical details

For two students, each completing a 30-credit thesis over 20 weeks. Work is carried out onsite at Bosch's Lund office. Preference is given to applicants based in the Lund and Malmö areas. Contact: Diego Medina, Diego.Medina@se.bosch.com.