positions

Task: Embedded active inference for autonomous robot manipulation (WP4)

Host institution: TU Delft

Country: Netherlands

Supervisor: Dr. K. Langendoen [TU Delft]

Co-supervisors: Prof. M. Wisse [TU Delft]; Dr. M. Ozo [XOsight]

Objectives: 1) To study how to design core structure for latent space of predictive models for active inference; 2) To investigate how to leverage encoder-decoder and attention structures for scalability; 3) To implement a hierarchical structure for further compression of model representation; 4) To optimise model structure and architecture of active inference algorithms to maximise edge device efficiency.

Expected Results: 1) A robust core structure for the latent space, enhancing predictive accuracy and model reliability in active inference. 2) A scalable, efficient model capable of handling large datasets with improved processing speed and accuracy. 3) A compressed model representation, reducing computational load and storage requirements while maintaining performance integrity and a refined model that significantly boosts the efficiency and effectiveness of active inference algorithms on edge devices.

PhD enrolment: Doctoral School of TU Delft

Planned secondments: 

• XOsight (3 months, M15-M17): Design of a scalable model to speed up large dataset processing in indoor robot inventory, with Dr. M. Ozo (KPI: joint paper)

• UMass (5 months, M26-M30): Compressed yet accurate on-device active inference model, with Prof. D. Ganesan (KPI: joint paper)

Candidate profile: computer science, robotics engineering, electrical engineering, telecommunication engineering (in order of preference)

Desirable skills/interests: machine learning, embedded systems/AI, robot operating system (ROS), signal processing, applied optimization (the applicant should be proficient in at least two of the skills)

Application Deadline: February 14, 2025, AoE

Descriptions of all the 18 DC Positions

Submit Your Application HERE