Dynamic Informed RRT* for TIAGo Service Robot Navigation in Hospital Environments: Preliminary Evaluation with Moving Obstacles

Dynamic Informed RRT* for TIAGo Service Robot Navigation in Hospital Environments: Preliminary Evaluation with Moving Obstacles 1st Alessandro Durat DEIB Politecnico di Milano Milano, Italy [email protected] 2nd Elena De Momi DEIB Politecnico di Milano Milano, Italy [email protected] 3rd Narcis Miguel Ba˜ nos PAL Robotics Barcelona, Spain [email protected] 4th Francesco Ferro PAL Robotics Barcelona, Spain [email protected] I. INTRODUCTION Modern hospitals are increasingly deploying autonomous mobile robots for surgical and ward support tasks, such as delivering sterile instruments, handling intra-room logistics, or assisting patients with limited mobility. Reliable navigation is crucial, yet clinical environments are highly cluttered and dynamic, with staff moving unpredictably in confined spaces. Path planners must therefore ensure safe, collision-free trajectories that adapt online to changing conditions. Classical grid-based algorithms, such as A* and Dijkstra, guarantee optimal paths on static maps but scale poorly in dynamic settings. Sampling-based methods like RRT* explore more efficiently and enable replanning, while Informed RRT* accelerates convergence by restricting sampling to an admissible ellipsoid [1] [2]. However, their systematic evaluation in hospital-like dynamic environments remains limited. The TIAGo platform (PAL Robotics), equipped with a human-aware mobile base, 3D LiDAR, RGB-D cameras, and a compliant manipulator, is well-suited for intra-facility logistics. This paper extends Informed RRT* to handle moving obstacles, evaluates its performance with dynamic personnel models, and assesses its suitability for real-time deployment on TIAGo. II. MATERIALS AND METHODS A. Simulated Maps A grid map, with each square measuring 0.1 m per side, was created from scratch to simulate a realistic, to-scale surgical room (7 × 9 m) (Fig. 1). The main obstacles, modeled to scale, include the surgical table (1), medical staff (2 and 3), and medical instrumentation such as the anesthesia station (4) and the instrument trolley (5). All obstacles were inflated to account for the dimensions of the TIAGo base configuration and to maintain a safety clearance around them, a strategy commonly adopted in mobile robot navigation to ensure safe operations [3]. This work was supported in part by PAL Robotics, Spain, and in part by the Italian Ministry of Universities and Research under PNRR ex DM 630, 2024. B. Path Planning Algorithms In the static benchmark, we compared the three pathplanning algorithms A*, Dijkstra, and Informed RRT* [1]: while A* and Dijkstra guarantee optimal solutions on grid maps, they generally incur higher computational costs. Informed RRT*, instead, explores the configuration space through random sampling and incremental tree construction. To improve convergence speed and path quality, the variant used in this work integrates not only elliptical sampling and tree pruning, but also an adaptive neighbourhood radius and a weighted goal bias: the adaptive radius decreases as nodes increase to balance connectivity and computational cost, while the goal bias accelerates path discovery without compromising the principle of random sampling. This combination yields faster feasible paths while preserving asymptotic optimality and broad space coverage, as also highlighted in goal-oriented RRT variants. For dynamic settings, Informed RRT* was adapted for moving obstacles: the robot executes the first feasible path as soon as available; if blocked, it halts, prunes the tree by redefining the ellipse between its current position and the goal, and replans within this region until a new valid path emerges. This execute–validate–replan cycle combines efficient exploration with incremental repair, making the planner suitable for crowded hospital environments. C. Experimental Protocol An extensive set of simulations was conducted to evaluate the three algorithms in the static scenario. Performance was assessed through the median values across trials, and at the end of each simulation the evaluation metrics defined in the following section were recorded. This process allowed us to identify the best candidate to be later adapted to the dynamic case. Similarly, to evaluate the dynamic version of Informed RRT*, we carried out three sets of 100 simulations each. In each set, the four obstacles representing medical staff (2 and 3, in Fig. 1) were made dynamic and assigned different movement patterns to test how robust the planner is under 2025 I-RIM Conference October 17-19, Rome, Italy ISBN: 9788894580570 10.5281/zenodo.17629796 165 Fig. 1. Evolution of the path generated by the dynamic Informed RRT* in the proposed hospital map. The orange segment indicates the portion already traversed by TIAGo before being blocked by a moving obstacle, while the green segment represents the replanned path from TIAGo’s stop position to the goal. different motion behaviors. All obstacles were inflated to account for TIAGo’s cargo dimensions and maintain a safety clearance. After each simulation, we recorded the evaluation metrics. Simulations used fixed start and goal positions across all runs, in order to isolate the effect of obstacle motion on planner performance and reduce external variation. The results from these experiments directly informed the decision to evolve Informed RRT* into its dynamic form, as it consistently showed superior performance under these controlled conditions. D. Evaluation Metrics In the static benchmark, we applied the following metrics, to determine the best performing method: path length, planning time, number of nodes, and turning-angle indicators. For the dynamic extension of Informed RRT*, where four moving obstacles were introduced, performance was evaluated using three additional metrics: the success rate reports the percentage of trials in which the robot successfully reached the goal without collision. The Dynamic suboptimality (Ldyn / LA*) measures the average executed path length (Ldyn) relative to the static A* optimum (LA*), thus indicating the cost of replanning in a dynamic environment and allowing fair comparison across different obstacle layouts, motion patterns, or map conditions. Finally, the Replanning latency quantifies the average time required to repair or regenerate a path whenever a dynamic obstacle blocks the current trajectory, measured per replanning event. These metrics reflect the criteria most commonly used to evaluate dynamic planners, including RRTX and D* Lite, and allow us to relate our results to trends established in the state of the art. E. Results Across a large number of static-map simulations, the proposed variant of Informed RRT* consistently outperformed A* and Dijkstra in terms of computation time to find the path, path length, and path smoothness, requiring approximately five times less computation time to obtain higher quality paths. In the dynamic scenario with four moving obstacles, preliminary results show a 100% success rate and a dynamic suboptimality of 1.20 relative to the static A* optimum. The measured replanning latency exceeded 100 ms, indicating room for optimization. These results are consistent with the ones reported in literature for other dynamic planners [4]: RRTX emphasizes fast incremental repair with near-optimal paths, and D* Lite achieves replanning latencies in the millisecond range while maintaining robustness. Our findings suggest that dynamic Informed RRT* offers competitive path quality and reliability, although further work is needed to reduce replanning latency to real-time levels. CONCLUSIONS AND DISCUSSION This work demonstrates that Informed RRT can be successfully extended to dynamic scenarios, achieving 100% success rate and competitive path quality, thus proving to be a promising path planning method for TIAGo in crowded surgical and hospital environments. REFERENCES [1] Y. Tang, M. A. Zakaria, and M. Younas, “Path planning trends for autonomous mobile robot navigation: A review,” Sensors, vol. 25, no. 4, 2025. [Online]. Available: https://www.mdpi.com/1424-8220/25/4/1206 [2] J. D. Gammell, S. S. Srinivasa, and T. D. Barfoot, “Informed rrt*: Optimal sampling-based path planning focused via direct sampling of an admissible ellipsoidal heuristic,” in 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, Sep. 2014, p. 2997–3004. [Online]. Available: http://dx.doi.org/10.1109/IROS.2014.6942976 [3] L. Liu, X. Wang, X. Yang, H. Liu, J. Li, and P. Wang, “Path planning techniques for mobile robots: Review and prospect,” Expert Systems with Applications, vol. 227, p. 120254, 2023. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S095741742300756X [4] M. Otte and E. Frazzoli, “Rrtx: Asymptotically optimal single-query sampling-based motion planning with quick replanning,” The International Journal of Robotics Research, vol. 35, 09 2015. 166