Bridging Probabilistic Inference and Behavior Trees: An Interactive Framework for Adaptive Multi-Robot Cooperation

TL;DR

The paper addresses the challenge of coordinating robot teams under partial observability by unifying probabilistic inference with Behavior Trees. It introduces the Interactive Inference Behavior Tree (IIBT), embedding free-energy–minimizing inference directly into BT execution nodes to enable online, decentralized policy selection and coordination. Key contributions include a formal multi-robot generative model, variational inference with ELBO, expected free energy–based policy selection, and a detailed IIBT node design that preserves BT interpretability while enabling belief updates and collaboration. Experimental results from simulations and real-world cooperative tasks demonstrate substantial reductions in BT complexity (up to 76% fewer nodes) while maintaining robust, adaptive coordination under uncertainty. The framework promises scalable, interpretable multi-robot cooperation and sets the stage for extensions to heterogeneous teams and online preference learning in complex environments.

Abstract

This paper proposes an Interactive Inference Behavior Tree (IIBT) framework that integrates behavior trees (BTs) with active inference under the free energy principle for distributed multi-robot decision-making. The proposed IIBT node extends conventional BTs with probabilistic reasoning, enabling online joint planning and execution across multiple robots. It remains fully compatible with standard BT architectures, allowing seamless integration into existing multi-robot control systems. Within this framework, multi-robot cooperation is formulated as a free-energy minimization process, where each robot dynamically updates its preference matrix based on perceptual inputs and peer intentions, thereby achieving adaptive coordination in partially observable and dynamic environments. The proposed approach is validated through both simulation and real-world experiments, including a multi-robot maze navigation and a collaborative manipulation task, compared against traditional BTs(https://youtu.be/KX_oT3IDTf4). Experimental results demonstrate that the IIBT framework reduces BT node complexity by over 70%, while maintaining robust, interpretable, and adaptive cooperative behavior under environmental uncertainty.

Figures (21)

• Figure 1: Overview of the proposed framework. Multiple robots perform interactive inference to jointly minimize free energy, dynamically update their policies, and coordinate actions in a shared, continuously evolving environment.
• Figure 2: The figure illustrates the interactive inference process between robots $\mathcal{R}_{i}$ and $\mathcal{R}_{j}$ using a generative model.
• Figure 3: Workflow of interactive nodes in the BT. Each robot collects its local observations $\mathcal{O}^{i}_{\tau}$, abstracts them into logical variables $\mathcal{L}^{i}$, and updates its belief $s^{i}_{\tau}$. The BT emits a preference matrix $\mathcal{C}^{i}$ to the inference module, which queries task models for $\mathcal{A}^{i}, \mathcal{B}^{i}_{\pi}, \mathcal{D}^{i}$, incorporates other robots’ intentions, and returns state/policy information back to the BT for execution.
• Figure 4: Two robots, $\mathcal{R}_{1}$ and $\mathcal{R}_{2}$, operate in a 7x7 grid, with cell positions labeled as $\{ p_{0}, p_{1}, \dots, p_{48} \}$. The environment features two goals, $goal_{1}$ and $goal_{2}$, and each robot creates paths to both goals.
• Figure 5: Fig.(a) shows the traditional method for robot $\mathcal{R}_{1}$ to select the nearest goal while considering other robots' states. Fig.(b) illustrates the interactive inference node, which selects a strategy to minimize free energy. Both figures demonstrate the same functionality.