Mind the Context: Continual Learning of Socially Appropriate Robot Actions via Environmental-Social Disentanglement

Noosaga Placements

• Human Robot Interactionsubfield90%
  The paper focuses on learning socially appropriate robot actions, which is a core topic in human-robot interaction, and evaluates on social scenarios with human annotations.
  Social robots are expected to operate across diverse environments, where similar arrangements can imply different socially appropriate actionslearning the appropriateness of robot actions (e.g., cleaning, serving, starting a conversation) across several indoor domains
• Classical Deliberative Roboticsframework80%
  The paper is situated within learning-based robotics, as it uses neural networks and continual learning to enable robots to learn appropriate actions.
  EDD uses panoptic segmentation to decompose each scenedual-branch network processes these inputs
• Machine Learningsubfield85%
  The core contribution is a continual learning method with replay-based rehearsal, addressing catastrophic forgetting, which is a machine learning problem.
  robots require continual learning (CL) to adapt from sequential experience while retaining previously acquired knowledgeuses replay-based rehearsal to mitigate forgetting
• Human-Centered and Participatory Designframework70%
  The work addresses social appropriateness in human-robot interaction, which is part of human-centered design considerations.
  socially appropriate robot actionshuman annotations that are used as ground-truth social appropriateness scores
• Roboticssubfield70%
  The work is applied to social robots and integrates perception and learning, fitting within the broader AI robotics subfield.
  Social robots are expected to operate across diverse environmentsEDD uses panoptic segmentation to decompose each scene
• Learning and Adaptationframework70%
  The paper focuses on continual learning and adaptation of robot behavior, which is a key aspect of learning and adaptation in HRI.
  continual learning (CL) to adapt from sequential experienceadapt their behaviour as environments change over time
• Robot Learningsubfield60%
  The paper presents a learning-based approach for robots to acquire action appropriateness, which aligns with robot learning.
  continual learning (CL) to adapt from sequential experiencelearning the appropriateness of robot actions

Abstract

Social robots are expected to operate across diverse environments, where similar arrangements can imply different socially appropriate actions, e.g., starting a conversation may be acceptable in a crowded home but disruptive in an office meeting. Because such norms and environments cannot all be anticipated in advance, robots require continual learning (CL) to adapt from sequential experience while retaining previously acquired knowledge. Prior work has studied CL for generating socially appropriate robot actions, but it has not addressed domain-incremental settings in which the robot incrementally encounters diverse contexts (e.g., living room, meeting room, office, hallway), where both environmental (e.g., whether the space is open or cluttered with furniture) and social cues (e.g., how people or other agents are positioned around the robot) jointly shape the appropriateness of robot actions. We address this gap with the Explicit Disentanglement Dual-Branch (EDD) framework. EDD explicitly separates environmental and social-agent related knowledge and uses replay-based rehearsal to mitigate forgetting while learning the appropriateness of robot actions (e.g., cleaning, serving, starting a conversation) across several indoor domains. Experiments show that EDD outperforms several state-of-the-art baselines, and ablation studies further evaluate different disentanglement strategies and the sensitivity to domain ordering. Our code is publicly available at https://github.com/Cambridge-AFAR/Mind-the-Context.git.

Paper Context