Visual Goal-Directed Meta-Learning with Contextual Planning Networks
This work addresses the challenge of enabling robots to quickly generalize to new tasks and goals, but it is incremental as it builds on existing meta-learning approaches with modest performance gains.
The paper tackled the problem of zero-shot goal-directed meta-learning for manipulation tasks by introducing contextual planning networks (CPN), which use goal images to condition the approach, and found that CPN outperformed other methods on one task and was competitive on others in evaluations across 24 Metaworld tasks.
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.