Course content

This course is offered every second year, next time in autumn 2027.It builds upon the foundational knowledge from TTK4255 Robotic Vision, with advanced focus on motion estimation, non-linear estimation methods, object recognition, semantic scene understanding, affordance detection, and high-level visual reasoning. Emphasis is placed on both theoretical underpinnings and practical implementation skills.The course is taught in English.

Special note on Artificial Intelligence (AI): AI is integrated into the course through learning-based methods for perception, including deep neural networks, transformer architectures, multimodal models, and embodied AI frameworks. Students analyze how AI-based techniques can be combined with classical geometric and statistical methods to achieve robust and explainable visual perception. Critical reflection is emphasized, including when AI methods are appropriate, what limitations they introduce, and how they affect robustness, uncertainty, and interpretability in advanced perception systems.

Learning outcome

Scientific Content (a selection of topics will be covered depending on participant interests and background):

• Optimization, estimation, and uncertainty models for robotic and visual perception systems
• Motion estimation, optical flow, and scene flow
• Scene and place recognition
• Dense, semi-dense, and sparse correspondence estimation
• Shape priors and shape estimation
• Pose estimation and tracking
• Semantic scene understanding and contextual reasoning
• Affordance detection and functional scene interpretation
• Statistical models and uncertainty quantification
• Segmentation and model fitting using probabilistic methods
• Correspondence and pose consistency
• Relational and structural recognition
• Learning-based approaches for perception, including deep representations, transformers, and multimodal models (vision-language models, embodied AI)

SKILLS:

• Apply linear and non-linear estimation methods to image and video data
• Design and evaluate advanced systems for motion estimation, object recognition, semantic understanding, and affordance reasoning
• Critically assess strengths and limitations of classical and AI-based methods and select appropriate techniques for complex perception tasks
• Integrate probabilistic, geometric, and learning-based tools into unified perception frameworks
• Use AI tools in a transparent, well-documented, and academically responsible manner

GENERAL COMPETENCE:

• Be able to apply the acquired knowledge to new areas and complete advanced research-oriented tasks
• Communicate extensive independent work, using precise terminology in advanced visual perception
• Contribute to innovative thinking and research development in visual perception and intelligent systems
• Demonstrate awareness of ethical challenges, data protection, and academic integrity when using AI tools in scientific work

Learning methods and activities

Study groups, selected readings, and optional problem sets. A mandatory course project with an oral presentation and written report.

Recommended previous knowledge

TTK4255 Robotic Vision

TTK4250 Sensor Fusion

Course materials

A collection of papers, which will be given at the beginning of the semester.