Visual Goal-Directed Meta-Learning with Contextual Planning Networks

TL;DR

This paper introduces contextual planning networks (CPN), a meta-learning approach that generalizes to new goals and tasks quickly, demonstrated on manipulation tasks with promising results and real-world robotic experiments.

Contribution

The paper presents CPN, a novel goal-conditioned meta-learning method that improves zero-shot generalization to new tasks and goals in manipulation settings.

Findings

CPN outperformed several baselines on one manipulation task.

CPN was competitive with existing approaches on other tasks.

Successful real-world demonstration on Jenga with a robotic arm.

Abstract

The goal of meta-learning is to generalize to new tasks and goals as quickly as possible. Ideally, we would like approaches that generalize to new goals and tasks on the first attempt. Toward that end, we introduce contextual planning networks (CPN). Tasks are represented as goal images and used to condition the approach. We evaluate CPN along with several other approaches adapted for zero-shot goal-directed meta-learning. We evaluate these approaches across 24 distinct manipulation tasks using Metaworld benchmark tasks. We found that CPN outperformed several approaches and baselines on one task and was competitive with existing approaches on others. We demonstrate the approach on a physical platform on Jenga tasks using a Kinova Jaco robotic arm.