[Paper Review] Hierarchical Haptic Manipulation for Complex Skill Learning

Simon Hangl, Emre Uğur+2 more|arXiv (Cornell University)|Mar 2, 2016

Robot Manipulation and Learning21 references3 citations

TL;DR

This paper proposes a hierarchical haptic manipulation framework that enables complex skill learning through autonomous skill playing and reusable basis skills. By transforming novel environments into known states via skill hierarchies, the method achieves generalization in complex dual-hand manipulation tasks with minimal supervision and without strong environmental assumptions.

ABSTRACT

In complex manipulation scenarios (e.g. tasks requiring complex interaction of two hands or in-hand manipulation), generaliza- tion is a hard problem. Current methods still either require a substantial amount of (supervised) training data and / or strong assumptions on both the environment and the task. In this paradigm, controllers solving these tasks tend to be complex. We propose a paradigm of maintaining simpler controllers solving the task in a small number of specific situations. In order to generalize to novel situations, the robot transforms the environment from novel situations to a situation where the solution of the task is already known. Our solution to this problem is to play with objects and use previously trained skills (basis skills). These skills can either be used for estimating or for changing the current state of the environment and are organized in skill hierarchies. The approach is evaluated in complex pick-and-place scenarios that involve complex manipulation. We further show that these skills can be learned by autonomous playing.

Motivation & Objective

• To address the challenge of generalization in complex manipulation tasks requiring multi-hand coordination or in-hand manipulation.
• To reduce reliance on large-scale supervised datasets and strong assumptions about the environment or task structure.
• To enable robots to generalize to novel situations by transforming them into known states using learned basis skills.
• To develop a scalable, hierarchical control framework that organizes skills for efficient task execution.
• To demonstrate that basis skills can be autonomously acquired through self-supervised playing.

Proposed method

• The approach organizes previously trained basis skills into a hierarchical structure to enable compositionality in complex tasks.
• The robot uses these basis skills to either estimate or actively change the current state of the environment.
• Novel situations are handled by transforming the environment into a state where a known skill can be applied.
• The system leverages haptic feedback to guide skill selection and state transformation during manipulation.
• Skills are learned autonomously through self-supervised exploration and playing with objects.
• The hierarchy enables modular control, where high-level decisions select appropriate low-level skills based on environmental state.

Experimental results

Research questions

• RQ1 How can robots generalize complex manipulation skills across novel environments without extensive retraining?
• RQ2 Can basis skills be learned autonomously through self-supervised playing?
• RQ3 How effective is a hierarchical skill organization in enabling generalization for dual-hand manipulation tasks?
• RQ4 To what extent can environment transformation via skill application improve task success in complex scenarios?
• RQ5 What role does haptic feedback play in enabling robust state estimation and skill selection?

Key findings

• The hierarchical haptic manipulation framework successfully generalizes to novel manipulation scenarios without requiring additional supervised data.
• Autonomous playing enables the acquisition of reusable basis skills that are effective in complex pick-and-place tasks.
• The method achieves reliable task execution in dual-hand manipulation by transforming novel states into known, solvable configurations.
• The use of skill hierarchies reduces controller complexity while maintaining performance in diverse environments.
• Haptic feedback enhances state estimation and enables accurate skill selection during task execution.
• The approach demonstrates robustness in complex scenarios where traditional methods fail due to lack of generalization.