Visual Goal-Directed Meta-Learning with Contextual Planning Networks
arXiv:2111.09908
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.
References in corpus (9)
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
- Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning
- Efficient Off-Policy Meta-Reinforcement Learning via Probabilistic Context Variables
- The Predictron: End-To-End Learning and Planning
- Universal Planning Networks
- Learning to Continually Learn
- TACO: Learning Task Decomposition via Temporal Alignment for Control
- Watch, Try, Learn: Meta-Learning from Demonstrations and Reward
- Hallucinative Topological Memory for Zero-Shot Visual Planning