Tour guide robot: a 5G-enabled robot museum guide

Stefano Rosa, Marco Randazzo, Ettore Landini, Stefano Bernagozzi, Giancarlo Sacco, Mara Piccinino, Lorenzo Natale

Abstract

This paper details the real-world deployment of a 5G-enabled humanoid robot, the R1, as a museum tour guide. By offloading heavy computational tasks—such as 3D obstacle avoidance and dialogue management—to remote servers via a 5G network, the researchers demonstrate a "cloud robotics" approach that allows for more capable, lightweight service robots. The study proves that 5G’s low latency and high bandwidth are critical for maintaining safety and interaction quality in dynamic, crowded environments.

Introduction

Service robots in public spaces like museums face a difficult trade-off. To navigate safely around crowds and interact naturally with visitors, they require significant onboard computing power. However, heavy hardware increases costs, generates heat, and drains batteries quickly.

While "cloud robotics" (moving computation to remote servers) is a potential solution, traditional wireless technologies like Wi-Fi and 4G often suffer from high latency and unstable connections. For a robot, a half-second delay in processing a "stop" command because of a network lag could result in a collision. The authors tackle the challenge of whether 5G can provide the reliable, high-speed link necessary to move safety-critical robotic "intelligence" off the robot and into the cloud.

System Architecture and Methods

The researchers utilized the R1 humanoid robot, developed by the Istituto Italiano di Tecnologia (IIT). The system architecture is a hybrid of the YARP middleware (for distributed communication) and ROS (for navigation).

The R1 humanoid robot hardware configuration, including sensors and the 5G smartphone used for connectivity.

Key technical approaches included:

  • Computational Offloading: While basic localization remained onboard for safety, the "heavy lifting"—including 3D obstacle avoidance (voxel-based ray-casting), path planning, crowd detection (using OpenPose), and the dialogue system—was moved to a remote server.
  • Network Optimization: To handle the asymmetric nature of 5G (where download speeds far exceed upload speeds), the team implemented aggressive sensor data compression (zlib for LiDAR/depth, MJPEG for RGB) and throttled sensor streams to 10Hz.
  • Behavior Trees (BTs): The high-level logic was managed via BTs, providing a reactive framework to handle both navigation tasks and network-related faults.
System architecture diagram showing the distribution of tasks between the robot (R1), remote servers, and external APIs.

Experimental Evaluation

The paper focuses on experimental evaluation in two distinct, real-world environments in Turin:

  • Galleria d’Arte Moderna (GAM): A large, feature-sparse environment (~60m x 20m).
  • Palazzo Madama: A smaller, more complex environment (~30m x 12m) filled with glass display cases.
  • User Feedback: A Quality of Experience (QoE) dataset was generated from surveys of over 100 visitors who interacted with the robot.

Key Insights

  • The Uplink Bottleneck: In robotics, the robot is the primary data producer. Most 5G commercial setups are optimized for consumer downloads. The authors found that network upload bandwidth is the actual "upper bound" for cloud robotics performance.
  • Stability over Speed: While 5G offers high peaks, its most valuable trait for robotics is the stability of the Round-Trip Time (RTT). Compared to Wi-Fi, which showed massive jitter (up to 500ms), 5G maintained a much tighter latency window (10-100ms), which is essential for closed-loop control.
  • Human-Robot Interaction (HRI) in the Wild: The authors noted that children were much more likely to crowd the robot. This necessitated the 5G-enabled 3D obstacle avoidance to detect small humans and limbs that 2D LiDARs might miss.

Results

  • Latency: The mean application-level latency was approximately 23.5 ms, well within the requirements for responsive navigation.
  • Throughput: Through compression and throttling, the team kept the robot's data footprint between 4–6 Mbps for upload and 15–20 Mbps for download.
  • Navigation Accuracy: The robot reached designated points of interest with a Root Mean Square Error (RMSE) between 0.07m and 0.15m.
  • Qualitative Success: Field trials lasted 10 days across two museums. Survey results showed high visitor engagement, particularly regarding the robot's ability to use natural gestures and facial expressions while narrating.

Limitations and Future Work

Despite the success, several hurdles remain. Speech recognition in noisy museum halls frequently caused the Google Dialogflow system to fail. Additionally, even with offloading, the R1’s battery only lasted about one hour. Future work will explore 5G "Network Slicing" to reserve dedicated bandwidth and the direct integration of 5G modems into the robot's chassis.

Summary updated

Abstract

p>This paper presents and discusses the development and deployment of a tour guide robot as part of the 5 g-TOURS EU research project, aimed at developing applications enabled by 5G technology in different use cases. The objective is the development of an autonomous robotic application where intelligence is off-loaded to a remote machine via 5G network, so as to lift most of the computational load from the robot itself. The application uses components that have been widely studied in robotics, (i.e., localization, mapping, planning, interaction). However, the characteristics of the network and interactions with visitors in the wild introduce specific problems which must be taken into account. The paper discusses in detail such problems, summarizing the main results achieved both from the methodological and the experimental standpoint, and is completed by the description of the general functional architecture of the whole system, including navigation and operational services. The software implementation is also publicly available.</p>
BibTeX
@article{rosa2023tour,
  author = {Stefano Rosa and Marco Randazzo and Ettore Landini and Stefano Bernagozzi and Giancarlo Sacco and Mara Piccinino and Lorenzo Natale},
  title = {{Tour guide robot: a 5G-enabled robot museum guide}},
  journal = {Frontiers Robotics AI},
  year = {2023},
  doi = {10.3389/frobt.2023.1323675},
}
Figure