Building an Interactive Map of Transfer Learning in Robotics

Context

Recent advances in robotics, exemplified by the commercial deployment of quadrupeds like Boston Dynamics’ Spot and Unitree’s Go, have demonstrated impressive capabilities in specific domains. However, these systems remain narrowly specialized – a robot trained for warehouse navigation cannot easily adapt to search-and-rescue missions, despite sharing underlying locomotion principles. Transfer learning addresses this fundamental limitation by enabling robots to leverage prior knowledge when learning new tasks, environments, or embodiments. The field centers on three critical questions: When to transfer? What to transfer? How to transfer? Despite growing research activity, the landscape remains fragmented, making it difficult for researchers to identify trends, gaps, and promising directions.

Figure 1: Visualize recent progress in transfer learning for robotics in interactive dashboard.

Project Overview

This project aims to create the first comprehensive, interactive map of transfer learning research in robotics (see figure 1). Building on the recent survey by Jaquier et al. (2024) [2], you will systematically analyze 5 years of top-tier robotics publications (CoRL, ICRA) to extract structured data on transfer learning approaches. The resulting dashboard will enable researchers to: (1) filter papers by transfer type (task/environment/embodiment), (2) track temporal trends in methodologies, (3) identify underexplored research areas, and (4) discover related work through interactive visualization. This tool will serve both the broader robotics community and inform LASA’s ongoing research directions in transfer learning.

Approach

The student will conduct a complete a literature review with
implementation of results in a dashboard:

• Tool selection and setup (Weeks 1-2): Evaluate visualization frameworks (Plotly Dash, Observable/D3.js, Streamlit), set up development environment, define data schema for paper extraction
• Literature review and database creation (Weeks 3-9): Systematically analyze CoRL (2021-2025) and ICRA (2021-2025) proceedings, extract structured metadata (transfer type, methods, benchmarks, results), build filterable database (≈100-150 papers)
•  Dashboard implementation (Weeks 10-12): Design and build interactive visualization with filtering, search, and trend analysis capabilities; iterate based on feedback from lab members
• Documentation and presentation (Week 13): Write final report following survey paper structure, prepare presentation with live demo

Tools: Python (Pandas, Plotly/dash or Streamlit), Zotero or similar reference manager, Notion or similar for database, Git/GitHub for version control. The final dashboard will be deployed as an open-access web application.

Student gain. Beyond developing deep expertise in transfer learning—one of robotics’ most active research areas—you will build a portfolio piece demonstrating both analytical and technical skills. The resulting dashboard may be featured on LASA’s website. You’ll also develop valuable ”research infrastructure” skills: systematic literature review, data extraction pipelines, and interactive web-based visualization – competencies increasingly important in modern AI/robotics research.

Prerequisites

Required:

• Strong analytical skills to extract key information from technical papers
• Python programming (pandas, data processing)
• Background in robotics/machine learning (e.g., completed relevant EPFL courses)

Preferred

• Experience with web frameworks or data visualization libraries (Plotly, D3.js)
• Familiarity with transfer learning concepts
• Git/GitHub for version control

References

A few references useful to understand the project

[1] Billard, A. et al. A roadmap for AI in robotics. Nat Mach Intell 7, 818–824 (2025).

[2] Jaquier, N. et al. Transfer learning in robotics: An upcoming breakthrough? A review of promises and challenges. The International Journal of Robotics Research 44, 465–485 (2025).