GIFT: Geometry-Induced Functional Transfer for Category-Level Object Manipulation
Cristiana de Farias, Luis Figueredo, Riddhiman Laha, Komlan Jean Maxime Adjigble, Brahim TAMADAZTE, Rustam Stolkin, Sami Haddadin, Naresh Marturi
AI summary
Problem
Robots struggle to generalize manipulation skills to unfamiliar objects or new environments because traditional trajectory and policy transfer methods fail when object shapes, contact points, or reference frames change significantly.
Approach
The framework extracts object-centric interaction functions from a single kinesthetic demonstration, transfers them to novel objects using Functional Map Correspondence, and generates constraint-preserving trajectories via constant-screw interpolation.
Key results
- Enables one-shot skill transfer to unseen category-level objects without additional training
- Preserves task-defining geometric and contact constraints by construction across shape variations
- Achieves consistent real-world task execution in cluttered scenes on a 7-DoF robot platform
- Provides an interpretable, geometry-first alternative to brittle end-to-end visuomotor policies
Why it matters
It provides a reliable, constraint-guaranteed approach for category-level robotic manipulation that generalizes across shape variations, making it highly relevant for real-world automation and human-robot collaboration.
Abstract
Robotic manipulation of unfamiliar objects in new environments is challenging due to limited generalisation capabilities. We propose a new skill transfer framework, GIFT (Geometry-Induced Functional Transfer), which enables a robot to transfer complex object manipulation skills and constraints from a single human demonstration. Our approach addresses the challenge of skill acquisition and task execution by deriving geometric representations from demonstrations focusing on object-centric interactions. By leveraging the Functional Maps (FMC) framework, we efficiently map interaction functions between objects and their environments, allowing the robot to replicate task operations across objects of similar topologies or categories, even when they have significantly different shapes. Additionally, our method incorporates screw interpo- lation (ScLERP) for generating smooth, geometrically-aware robot paths to ensure the transferred skills adhere to the demonstrated task constraints. We validate the effectiveness and adaptability of our approach through extensive experiments, demonstrating successful skill transfer and task execution in diverse real-world environments without requiring additional training.