Core Concepts
MAkEable is a versatile framework that enables the transfer of mobile manipulation skills across tasks, environments, and robots through memory-centric and affordance-based approaches.
Abstract
The MAkEable framework focuses on facilitating the transfer of mobile manipulation skills by integrating affordance-based task descriptions into a memory-centric cognitive architecture. The framework allows for the autonomous execution of uni- and multi-manual manipulation actions across different robotic platforms. Real-world experiments demonstrate its applicability in grasping known and unknown objects, object placing, bimanual object grasping, skill transfer between humanoid robots, and pouring tasks learned from human demonstrations.
I. INTRODUCTION
Efficient transfer of learned tasks crucial in robotics.
MAkEable facilitates transfer of capabilities across tasks, environments, robots.
II. RELATED WORK
Different levels of task descriptions for robotic mobile manipulation.
III. THE FRAMEWORK
Design principles focus on modularity, extensibility, interpretability.
Task description based on affordances using IDF format.
System architecture involves discovery, parameterization, validation, selection, execution steps.
IV. USE CASES
Experiments include table-clearing with known/unknown objects, bimanual grasping tasks on ARMAR-6 robot, memory-enabled drawer-opening skill transfer between ARMAR-6 and ARMAR-DE humanoid robots.
V. EXPERIMENTS
Demonstrates real-world scenarios showcasing the versatility and adaptability of the MAkEable framework in various mobile manipulation tasks.
Stats
"Our framework integrates an affordance-based task description into the memory-centric cognitive architecture."
"Demonstrate the applicability of the framework in real-world experiments for multiple robots."
"ARMAR humanoid robots equipped with anthropomorphic arms used in experiments."