[Paper Review] Norms, Institutions, and Robots

Stevan Tomic, Federico Pecora+1 more|arXiv (Cornell University)|Jul 30, 2018

Multi-Agent Systems and Negotiation54 references3 citations

TL;DR

This paper proposes a formal framework that uses 'institutions' to represent social norms in mixed human-robot societies, enabling robots to reason about norms through a computational model grounded in constraint satisfaction problems (CSP). The framework supports norm verification, planning, and execution across physical domains, demonstrating successful real-world robot-human interactions in trading scenarios with artifacts like batteries and sponges.

ABSTRACT

Interactions within human societies are usually regulated by social norms. If robots are to be accepted into human society, it is essential that they are aware of and capable of reasoning about social norms. In this paper, we focus on how to represent social norms in societies with humans and robots, and how artificial agents such as robots can reason about social norms in order to plan appropriate behavior. We use the notion of institution as a way to formally define and encapsulate norms, and we provide a formal framework for institutions. Our framework borrows ideas from the field of multi-agent systems to define abstract normative models, and ideas from the field of robotics to define physical executions as state-space trajectories. By bridging the two in a common model, our framework allows us to use the same abstract institution across physical domains and agent types. We then make our framework computational via a reduction to CSP and show experiments where this reduction is used for norm verification, planning, and plan execution in a domain including a mixture of humans and robots.

Motivation & Objective

• To address the lack of operational semantics and physical grounding in existing normative models for robots.
• To enable robots to reason about social norms in mixed human-robot societies by formally linking abstract norms to physical domains.
• To support artifact-based coordination, where robots and humans interact via shared objects such as batteries or sponges.
• To provide a reusable, domain-agnostic normative model applicable across different agent types and social contexts.
• To bridge multi-agent systems and robotics by integrating abstract normative modeling with physical execution trajectories.

Proposed method

• The framework defines 'institutions' as formal abstractions that encapsulate roles, actions, artifacts, and norms in a social context.
• It introduces a grounding mechanism that maps abstract institutional elements to concrete physical agents and objects in a domain.
• The framework reduces normative reasoning tasks to constraint satisfaction problems (CSP), enabling computational verification and planning.
• Physical executions are modeled as state-space trajectories, linking abstract norms to observable behaviors over time.
• A timeline-based planner uses the CSP-encoded norms to generate compliant action sequences for robots.
• The system supports dynamic reconfiguration by changing role and artifact grounding, enabling reuse across different human-robot and robot-robot scenarios.

Experimental results

Research questions

• RQ1 How can social norms be formally represented and encapsulated in a way that is reusable across different physical domains and agent types?
• RQ2 How can abstract normative models be grounded to physical environments to ensure operational semantics for robotic execution?
• RQ3 Can constraint satisfaction problems (CSP) be effectively used to verify, plan, and execute norm-compliant behaviors in mixed human-robot interactions?
• RQ4 How can artifacts—physical objects used in interactions—be formally integrated into normative models for robotic coordination?
• RQ5 What is the feasibility of using the same institutional model across diverse scenarios involving humans, robots, or mixed teams?

Key findings

• The framework successfully verified norm compliance in a human-robot trading interaction, identifying a trajectory where a human gave a yellow sponge before picking a blue one as non-adherent due to violation of a temporal norm.
• A compliant plan was automatically generated and executed by a Pepper robot in a reconfigured scenario, where the robot played the buyer role and interacted with a human seller using a battery as the goods.
• The same institutional model was reused in a robot-robot interaction by changing the role grounding, demonstrating the framework’s reusability across agent types.
• The CSP-based reduction enabled efficient norm verification and planning, with the system detecting non-adherent behaviors through constraint violation analysis.
• The framework supported artifact-aware coordination, as demonstrated by the use of sponges and batteries as physical artifacts in the interaction model.
• The approach enabled seamless transition from abstract norm specification to physical robot execution, validated in real-world experiments with human participants and robots.