Adapting Manipulation Datasets to Humanoid Robots
Retargeting robot demonstrations to humanoid platforms and validating adapted datasets in simulation for imitation learning.
Data Science and Engineering2026–2027 · Current edition
Applied Data
Science Project.
From AI concepts to
real-world solutions.
Learn to innovate with AI in digital and physical environments, with people guiding the design and management supporting every step.
The project mindset · Four foundations
01
Design
User needs & requirements
02
Develop
Data, models & validation
03
Communicate
Results & impact
ManageAcross every stage
Planning · Coordination · Resources · KPIs · Risks
Design, develop and communicate. Manage throughout.
- Course codes
- 02TXXWS
02TXXSM - Course credits
- 8 CFU
- Lectures
- 30 hours
- Laboratory
- 50 hours
- Teaching language
- English
- When
- 1st semester · Year 2
The courseLearn by building
Learn by building
innovative AI solutions.
The course is offered in English in the first semester of the 2nd year of Data Science and Engineering at Politecnico di Torino.
Learn to turn ideas into innovative AI applications for digital and physical environments. Start with people’s needs, explore what emerging AI models make possible, and design, develop and validate a solution through a team project. Human-centred design, stakeholder mapping, user personas and functional requirements connect technological opportunities with the people who will use the solution.
The project mindset has four foundations: Design, Develop, Communicate and Manage. Design frames user needs and requirements; Develop brings together data, models, implementation and validation; Communicate explains the results through technical and scientific communication. Manage supports all three through innovation management, planning, coordination, key performance indicators (KPIs) and risk management.
The course combines 30 hours of lectures with 50 hours of laboratory activities. Explore language, vision and multimodal models, retrieval-augmented generation, transfer learning, AI ethics and the Sustainable Development Goals, alongside an introduction to vision-language-action models and embodied AI. Apply these concepts through practical case studies, team project development and presentation checkpoints.
The learning path
Four foundations. One project.
Design, develop and communicate.
Manage supports the whole journey.
01
Design
Start with people.Understand user needs, define the problem, identify stakeholders and set clear requirements.
Explore lectures02
Develop
Build and validate.Work with data and models, implement the solution, then test and validate it against the requirements.
Explore lectures03
Communicate
Make the impact clear.Explain the results and their impact through clear technical and scientific communication.
Explore lectures04
Across every stageManage
Support every stage.Plan the project, coordinate people and resources, track key performance indicators (KPIs) and manage risks across Design, Develop and Communicate.
Explore lecturesExplore & apply
Your course toolkit.
30 hours of lectures
50 hours of hands-on laboratory
Lecture topics are listed below. Slides will appear here as they become available.
Lectures & project tools16 topics
-
L01
Introduction
Course objectives, activities and the applied data science project.
Material pending -
L02
Project mindset
Design, develop and communicate, with manage supporting the whole project.
Material pendingRecommended papers 2
- Guidelines for Human-AI Interaction
Connects human needs, feedback and responsible AI behaviour to the Design foundation.
- Hidden Technical Debt in Machine Learning Systems
Explains dependencies, feedback loops and maintainability across Develop and Manage.
- Guidelines for Human-AI Interaction
-
03
Human-centred design, stakeholder mapping, user personas and functional requirements
Material pending -
04
Management, WBS and Gantt
Material pendingRecommended papers 2
- Project management: cost, time and quality, two best guesses and a phenomenon, its time to accept other success criteria
Connects project success with stakeholder benefits and performance indicators beyond cost, time and quality.
- Earned Value Project Management Method and Extensions
Connects planning with monitoring, forecasting and corrective action through earned value.
- Project management: cost, time and quality, two best guesses and a phenomenon, its time to accept other success criteria
-
05
Model-centric and data-centric projects
Material pendingRecommended papers 2
- Data-centric Artificial Intelligence: A Survey
Surveys training-data development, inference data and data maintenance across the AI lifecycle.
- "Everyone wants to do the model work, not the data work": Data Cascades in High-Stakes AI
Shows how data quality issues propagate through AI development and affect downstream outcomes.
- Data-centric Artificial Intelligence: A Survey
-
06
Foundation models and model hub
Material pendingRecommended papers 4
- On the Opportunities and Risks of Foundation Models
Introduces the foundation-model framing, covering capabilities, adaptation, opportunities and risks.
- Scaling Laws for Neural Language Models
Relates language-model performance to model size, dataset size and training compute.
- Transformers: State-of-the-Art Natural Language Processing
Describes the Transformers library and the reuse of pretrained language models.
- Model Cards for Model Reporting
Supports model selection through documentation of intended uses, evaluation and limitations.
- On the Opportunities and Risks of Foundation Models
-
07
Large language models and applications (LLMs)
Material pendingRecommended papers 2
- Attention Is All You Need
Provides the Transformer and attention mechanisms underlying modern language models.
- Language Models are Few-Shot Learners
Connects autoregressive language modelling with few-shot learning across diverse tasks.
- Attention Is All You Need
-
08
Transfer learning and domain adaptation
Material pendingRecommended papers 2
- A Survey on Transfer Learning
Clarifies source and target domains and tasks, and the relationship between transfer learning and domain adaptation.
- LoRA: Low-Rank Adaptation of Large Language Models
Explains adaptation through small low-rank updates while pretrained weights remain frozen.
- A Survey on Transfer Learning
-
09
Retrieval-augmented generation (RAG)
Material pending -
10
Vision models and applications
Material pending -
11
Vision-language models and applications
Material pending -
12
Transfer learning with vision-language models
Material pending -
13
Multimodal large language models and applications
Material pending -
14
AI ethics and Sustainable Development Goals (SDGs)
Material pending -
15
Introduction to AI in physical environments: vision-language-action models and embodied AI
Material pending -
16
Technical and scientific communication
Material pending
Try another topic.
Adjust your search or choose another foundation.
Put it into practice
In the laboratory.
Work with people, models and data.
Then bring them together in your team project.
- Hands-on practical case studies: design and development
- Team project development
- Team presentation checkpoints
For your team project
Ideas put to work.
Explore project proposals for 2026–2027.
Find a challenge to develop with your team.
Predicting Vision-Language-Action Model Failures
Predicting robot policy failures from early episode observations to support timely intervention.
Understanding Vision-Language-Action Model Failures
Stress-testing robot policies and relating failure patterns to gaps in training-data coverage.
Enterprise AI Runtime Optimization
Improving prompt reusability, response efficiency and document extraction in enterprise AI workflows.
Foundation Models for High-Resolution Air Quality Maps
Benchmarking and adapting foundation models to estimate pollutant concentrations at higher spatial resolution.
Artwork Type Classification for Metadata Suggestion
Classifying artefact types from images to support curators and improve artwork metadata suggestions.
Content Safety Classification for Generative AI Systems
Classifying prompts and model responses by safety and harm category, with fine-tuning and evaluation of guardrail models.
DEEP: Dupilumab Eosinophilia Early Prediction
Exploring clinical and demographic features to identify the risk of elevated blood eosinophils during Dupilumab treatment.
Geo-Embeddings for Burned Area Delineation
Assessing geo-embeddings as prior information for segmenting wildfire burned areas in satellite imagery.
Shot Recognition in Padel Match Videos
Detecting when shots occur and classifying their types from padel videos, player poses and ball trajectories.
Synthetic Dataset Generation for Car-Wash Hazard Detection
Generating, validating and labelling synthetic images of car-wash hazards to build a dataset for future detector models.
Snow Height Forecasting and Avalanche Risk Estimation
Forecasting snow height in Valle d’Aosta, with an extension to estimating avalanche risk using weather data.
Efficient Adaptation of Multimodal Models for Visual Storytelling
Adapting multimodal language models with parameter-efficient fine-tuning to generate coherent stories from image sequences.
Wildfire Smoke Segmentation from Space
Segmenting smoke from geostationary satellite images and evaluating transfer from US GOES data to European Meteosat-12 imagery.
Dear LLM, Can You Explain My MongoDB Database?
Reconstructing MongoDB collections, relationships and modelling patterns with LLMs, and grounding explanations in schema and document evidence.
Before you begin
Come prepared. Stay curious.
The background knowledge and reading
to support your project.
Useful foundations.
- Statistics
- Data mining
- Machine learning and deep learning
- Python language
- Relational, NoSQL, graph databases
Course codes 02TXXWS · 02TXXSM · 2nd year
Data Science and Engineering
Keep learning.
Recommended research to explore alongside the lectures.
Recommended papers
- Guidelines for Human-AI Interaction Project mindset
- Hidden Technical Debt in Machine Learning Systems Project mindset
- Project management: cost, time and quality, two best guesses and a phenomenon, its time to accept other success criteria Management, WBS and Gantt
- Earned Value Project Management Method and Extensions Management, WBS and Gantt
- Data-centric Artificial Intelligence: A Survey Model-centric and data-centric projects
- "Everyone wants to do the model work, not the data work": Data Cascades in High-Stakes AI Model-centric and data-centric projects
- On the Opportunities and Risks of Foundation Models Foundation models and model hub
- Scaling Laws for Neural Language Models Foundation models and model hub
- Transformers: State-of-the-Art Natural Language Processing Foundation models and model hub
- Model Cards for Model Reporting Foundation models and model hub
- Attention Is All You Need Large language models and applications (LLMs)
- Language Models are Few-Shot Learners Large language models and applications (LLMs)
- A Survey on Transfer Learning Transfer learning and domain adaptation
- LoRA: Low-Rank Adaptation of Large Language Models Transfer learning and domain adaptation
Reference books and reading materials
- Machine Learning Yearning, by Andrew Ng
- Data Science from Scratch, Joel Grus
- Harvard Business Review Project Management Handbook: How to Launch, Lead, and Sponsor Successful Projects, by Antonio Nieto-Rodriguez
- Oxford Guide to Effective Writing and Speaking: How to Communicate Clearly, by John Seely
- The Design of Everyday Things: Revised and Expanded Edition, by Donald Norman
- Noessel C. Designing Agentive Technology. AI That Works for People. Rosenfeld, 2013
- Proposal for a Regulation laying down harmonised rules on artificial intelligence, European Commission, 2021
Teaching materials are available through this website and the teaching portal.
How the course works
The course combines 30 hours of lectures with 50 hours of laboratory activities. Lectures connect the four project foundations with AI methods and applications in digital and physical environments. Laboratory work includes hands-on case studies in design and development, team project development and team presentation checkpoints.
Learn together
Your teaching team.
The people supporting
the 2026–2027 edition.
Giuseppe Rizzo
Course lead · AI & data science
Antonella Frisiello
Service design & UX
Alessandro Fiori
Teaching assistant
Federico D’Asaro
Teaching assistant
Luca Barco
Teaching assistant