Causality-enhanced Decision-Making for Autonomous Mobile Robots in Dynamic Environments

TL;DR

This work addresses autonomous robots operating in dynamic human environments by introducing a causality-enhanced decision-making framework that combines causal discovery from sensor data with causal reasoning for motion planning. The framework learns a context-aware causal model and uses do-calculus-based inferences to estimate battery consumption $\hat{L}$ and people density $\hat{D}$, integrating these quantities into an A*-based planner via a latency-stable heuristic. Key contributions include the end-to-end ROS-integrated architecture, the Gazebo-based PeopleFlow simulator for context-sensitive HRSI, and a thorough warehouse-case evaluation showing substantial gains in task success (≈89% vs 56%), safety (fewer collisions and better proxemic adherence), and energy efficiency, all while maintaining real-time planning costs (≈0.32 s per query). The results demonstrate that causality-aware planning can improve both efficiency and safety in human-shared environments, with the potential for long-term autonomy through online model updates and multi-robot coordination. Future work will validate the framework on real robots, enable autonomous model lifecycle management, and scale to multi-robot fleets.

Abstract

The growing integration of robots in shared environments - such as warehouses, shopping centres, and hospitals - demands a deep understanding of the underlying dynamics and human behaviours, including how, when, and where individuals engage in various activities and interactions. This knowledge goes beyond simple correlation studies and requires a more comprehensive causal analysis. By leveraging causal inference to model cause-and-effect relationships, we can better anticipate critical environmental factors and enable autonomous robots to plan and execute tasks more effectively. To this end, we propose a novel causality-based decision-making framework that reasons over a learned causal model to assist the robot in deciding when and how to complete a given task. In the examined use case - i.e., a warehouse shared with people - we exploit the causal model to estimate battery usage and human obstructions as factors influencing the robot's task execution. This reasoning framework supports the robot in making informed decisions about task timing and strategy. To achieve this, we developed also PeopleFlow, a new Gazebo-based simulator designed to model context-sensitive human-robot spatial interactions in shared workspaces. PeopleFlow features realistic human and robot trajectories influenced by contextual factors such as time, environment layout, and robot state, and can simulate a large number of agents. While the simulator is general-purpose, in this paper we focus on a warehouse-like environment as a case study, where we conduct an extensive evaluation benchmarking our causal approach against a non-causal baseline. Our findings demonstrate the efficacy of the proposed solutions, highlighting how causal reasoning enables autonomous robots to operate more efficiently and safely in dynamic environments shared with humans.

Figures (13)

• Figure 1: A mobile robot reasoning on the causal model of human spatial behaviours in a warehouse environment to navigate safely and efficiently.
• Figure 2: Block scheme of the causality-based decision making framework, consisting of three main pipelines: (i) Data Extraction which gathers and preprocesses data from the observed scenario; (ii) Learning that retrieves the causal model describing the observed scenario and exploits its structure along with extracted data to learn the parameters of the causal inference engine; (iii) Inference, which uses the learnt causal inference engine to estimate relevant quantities for determining a path to complete a task and deciding in advance whether execute it or abort it.
• Figure 3: Overview of PeopleFlow’s context-aware agent behaviour strategy. The Context Manager node handles the contextual factors—in this example, time. After the agent picks up a box from a shelf, it requests a new task. The Context-Pedsim Bridge node receives the task request and generates a new task for the agent depending on the current context. Since it is working-time, the agent is tasked to bring the box to the delivery point. A corresponding sequence of waypoints is generated and passed to Pedsim, which navigates the agent to the goal using the social force model. The agent’s goal is marked with a blue circle, while white circles represent the remaining waypoints.
• Figure 4: (a) An illustrative example of an Amazon warehouse setting. (b) Warehouse-like scenario map with obstacles depicted as black-band rectangles, and robot target stations as diamonds. Its implementation in our simulator: (c) Gazebo view, (d) RViz view.
• Figure 5: Causal model of the scenario staged by PeopleFlow. Contextual factors $W$, $S$, $C$, and $O$ are shown in grey, while system variables $V$, $L$, and $D$ are highlighted in orange.