CAMETA: Conflict-Aware Multi-Agent Estimated Time of Arrival Prediction for Mobile Robots

TL;DR

CAMETA addresses the challenge of predicting arrival times for multiple agents operating in unstructured indoor environments by introducing a three-layer framework: path planning, multi-agent ETA prediction on a spatio-temporal heterogeneous graph, and time-constrained path selection. The multi-agent ETA predictor uses a novel heterogeneous graph representation and a HEAT-based GNN to forecast per-agent ETAs while accounting for conflicts and imperfect plan execution. The approach yields substantial improvements in ETA accuracy, with $ ext{MAPE}$ improvements of 29.5% and 44% over naive baselines when comparing IMS and DMS variants, respectively, and demonstrates robustness to noise and varying robot densities. Although CAMETA does not guarantee conflict-free paths, it effectively reduces conflicts by informing a local planner and enabling more scalable, resilient multi-robot coordination in indoor settings without reliance on fixed infrastructure.

Abstract

This study presents the conflict-aware multi-agent estimated time of arrival (CAMETA) framework, a novel approach for predicting the arrival times of multiple agents in unstructured environments without predefined road infrastructure. The CAMETA framework consists of three components: a path planning layer generating potential path suggestions, a multi-agent ETA prediction layer predicting the arrival times for all agents based on the paths, and lastly, a path selection layer that calculates the accumulated cost and selects the best path. The novelty of the CAMETA framework lies in the heterogeneous map representation and the heterogeneous graph neural network architecture. As a result of the proposed novel structure, CAMETA improves the generalization capability compared to the state-of-the-art methods that rely on structured road infrastructure and historical data. The simulation results demonstrate the efficiency and efficacy of the multi-agent ETA prediction layer, with a mean average percentage error improvement of 29.5% and 44% when compared to a traditional path planning method (A *) which does not consider conflicts. The performance of the CAMETA framework shows significant improvements in terms of robustness to noise and conflicts as well as determining proficient routes compared to state-of-the-art multi-agent path planners.

Figures (5)

• Figure 1: The conflict-aware multi-agent planner framework is depicted in three layers. The first layer generates multiple path suggestions for a single agent. The second layer, enclosed in blue borders, produces an estimated time of arrival prediction for all agents. Finally, the third layer computes the overall cost for each suggested path, using the information generated in the previous layer.
• Figure 2: Illustration of the proposed multi-agent ETA prediction layer. A simplified illustration of the three layers representing the indoor environment and the connection between the layers is shown on the left. The three layers consist of: a 2D occupancy grid map of the environment, a static layer containing all nodes and edges with static information, and a dynamic layer containing the time-variant features of the graph. The proposed GNN architecture for multi-agent ETA prediction is also shown.
• Figure 3: The figure illustrates several examples of the maps utilized in training the multi-agent ETA prediction module (enclosed in red), and the map utilized for evaluating the overall performance of the system (enclosed in blue). The depiction of the black pixels indicates occupied spaces, while the white pixels signify unoccupied areas.
• Figure 4: The plot shows the increase in the avg. makespan as noise is applied in various degrees. As the cost function prioritizes the agents with the tightest schedule, a stable makespan can be seen for CAMETA.
• Figure 5: The plot shows the increase in avg. sum of costs as noise is applied in various degrees. Both PIBT and CBS rises as the noise increases, while $A^*$ and CAMETA is less affected by the noise.