Student project: Physical AI for Edge Robotics: Sparse Multimodal Learning and Sensor Fusion

Design efficient, sparse, multimodal AI models that enable real-time intelligence on resource-constrained robotic systems.

You will design and develop efficient multimodal AI models for robotics applications, structured as follows:

• Review state-of-the-art Physical AI models for robotics, including multimodal fusion and edge-efficient architectures (3 weeks)
• Study and compare emerging paradigms: attention-based models, state-space models (SSMs), and sparse neural networks (4 weeks)
• Design a novel multimodal AI architecture integrating heterogeneous sensor data (vision, radar, etc.) (7 weeks)
• Develop and train models using efficient learning strategies (e.g., training, pruning, token reduction) (6 weeks)
• Implement edge-aware optimization techniques (latency, memory, and energy constraints) (4 weeks)
• Evaluate performance on robotics-oriented datasets (perception, navigation, etc.) (4 weeks)
• Investigate fusion strategies: early, late, and cooperative multimodal fusion (3 weeks)
• Demonstrate the model on embedded or edge platforms (3 weeks)
• Refine architecture based on accuracy-efficiency trade-offs (2 weeks)
• Prepare thesis, documentation, and final presentation (3 weeks)

• Gain hands-on experience with multimodal sensing systems and real-world robotic data in the domain of Physical AI
• Explore cutting-edge paradigms including state-space models, attention mechanisms, and sparse AI
• Contribute to hardware-efficient AI co-design, aligning with imec’s research in edge computing
• Opportunity to publish results in top-tier AI, robotics, or systems conferences

• Collaborate with imec experts in sensor fusion, neuromorphic computing, and edge AI

• Master’s student in Artificial Intelligence, Computer Science, or related field
• Strong interest in the development of AI models, and multimodal learning
• Solid programming skills in Python and PyTorch (or similar frameworks)
• Familiarity with at least one of the following is a plus: (a) Transformer-based models or attention mechanisms, (b) State-space models (SSMs), (c) Sparse neural networks, (d) Multimodal learning or sensor fusion
• Experience with robotics or sensor data is a plus
• Ability to work independently and in a research team
• Strong analytical and problem-solving skills
• Good communication skills in English
• Must have the legal right to undertake an internship or thesis in the Netherlands.
• Must be available for a 9-month period, including time for thesis writing

Does this position sound like an interesting next step in your career at imec? Don’t hesitate to submit your application by clicking on ‘APPLY’.