A General Purpose Supervisory Signal for Embodied Agents
This addresses the challenge of ensuring self-supervised objectives encode task-relevant information for embodied agents, offering a generally applicable and simple-to-implement solution.
The paper tackles the problem of training embodied AI agents without manual reward engineering by proposing the Scene Graph Contrastive (SGC) loss, which uses scene graphs as a supervisory signal to align agent representations with environmental encodings, resulting in significant gains on tasks like Object Navigation, Multi-Object Navigation, and Arm Point Navigation.
Training effective embodied AI agents often involves manual reward engineering, expert imitation, specialized components such as maps, or leveraging additional sensors for depth and localization. Another approach is to use neural architectures alongside self-supervised objectives which encourage better representation learning. In practice, there are few guarantees that these self-supervised objectives encode task-relevant information. We propose the Scene Graph Contrastive (SGC) loss, which uses scene graphs as general-purpose, training-only, supervisory signals. The SGC loss does away with explicit graph decoding and instead uses contrastive learning to align an agent's representation with a rich graphical encoding of its environment. The SGC loss is generally applicable, simple to implement, and encourages representations that encode objects' semantics, relationships, and history. Using the SGC loss, we attain significant gains on three embodied tasks: Object Navigation, Multi-Object Navigation, and Arm Point Navigation. Finally, we present studies and analyses which demonstrate the ability of our trained representation to encode semantic cues about the environment.