Human-Centered Development of Guide Dog Robots: Quiet and Stable Locomotion Control

TL;DR

The paper tackles the challenge of making quadruped guide-dog robots acceptable to BLV users by prioritizing noise suppression and stable, human-like locomotion. It adopts a human-centered development process comprising exploratory stakeholder research, a novel NMPC+WBIC control framework with real-time SQP updates and a perception-based stair-climbing module, and thorough indoor user evaluations. The proposed controller reduces walking noise by up to $10$ dB (roughly halving perceived noise) and enables smooth, fast walking and comfortable stair navigation, validated through hardware experiments and a four-part BLV user study. The work demonstrates the feasibility and user-perceived benefits of quiet, stable quadruped locomotion for BLV mobility aids, offering a practical path toward deploying guide-dog robots in real-world environments while outlining limitations and avenues for future improvement.

Abstract

A quadruped robot is a promising system that can offer assistance comparable to that of dog guides due to its similar form factor. However, various challenges remain in making these robots a reliable option for blind and low-vision (BLV) individuals. Among these challenges, noise and jerky motion during walking are critical drawbacks of existing quadruped robots. While these issues have largely been overlooked in guide dog robot research, our interviews with guide dog handlers and trainers revealed that acoustic and physical disturbances can be particularly disruptive for BLV individuals, who rely heavily on environmental sounds for navigation. To address these issues, we developed a novel walking controller for slow stepping and smooth foot swing/contact while maintaining human walking speed, as well as robust and stable balance control. The controller integrates with a perception system to facilitate locomotion over non-flat terrains, such as stairs. Our controller was extensively tested on the Unitree Go1 robot and, when compared with other control methods, demonstrated significant noise reduction -- half of the default locomotion controller. In this study, we adopt a mixed-methods approach to evaluate its usability with BLV individuals. In our indoor walking experiments, participants compared our controller to the robot's default controller. Results demonstrated superior acceptance of our controller, highlighting its potential to improve the user experience of guide dog robots. Video demonstration (best viewed with audio) available at: https://youtu.be/8-pz_8Hqe6s.

Figures (9)

• Figure 1: Human-centered Development of a Guide-dog Robot. We employed a human-centered approach to develop a guide-dog robot’s locomotion controller, consisting of: 1) an initial exploratory study to identify critical unmet needs, 2) system (technology) development, and 3) user evaluation. One of our major contributions is the adoption of this complete, user-centric development process to advance the robot controller for blind and low-vision individuals.
• Figure 2: Overall Control Framework. Our guide-dog controller uses an NMPC+WBIC architecture. The NMPC incorporates a single-rigid-body model with full orientation dynamics to enable better orientation control and computes the required reaction forces, which are then passed to the WBIC. The WBIC generates feedforward torques, as well as desired joint positions and velocities, which are subsequently fed into the robot’s onboard motor controller. For stair climbing, the robot is equipped with a RealSense D435 camera, and the captured point clouds are processed into a height map using Elevation Mapping.
• Figure 3: Comparison of Our MPC and Convex MPC. We compare our controller with the convex MPC + WBIC framework presented in kim2019highly to evaluate the improved balance stability resulting from our new MPC formulation. In the Mujoco simulation, we command the robot to move forward at $1.2~m\per s$ while applying a $25~N$ pulling force to simulate the force exerted by a handler. During walking, we apply an impulse force of $100~N$ for 0.01 seconds to assess the controllers' robustness. The tests were performed at three different gait speeds (i.e., swing times of 0.2, 0.25, and 0.3 seconds). The results show that, in both the original implementation and the new implementation using OSQP, convex MPC exhibits limited capability in sustaining balanced walking compared to ours.
• Figure 4: Hardware Experiment Results. (a) To accurately measure the noise BLV individuals hear during walking, a person records the decibel value at his ear level while following the robot. (b) Both controllers (Our vs Default) are traveling at 1 m/s. The slower gait requires our controller to travel more distance within a single step. (c) On average, our controller has achieve a 10 decibel reduction. (d) We observed that angular velocity amplitudes of each controller are similar or even less sometimes. This shows our approach maintains comparable or better orientation control while having slower gait frequency. (e) Our controller shows better performance in maintaining the body velocity.
• Figure 5: Stair Climbing Tests. Comparison of the default and proposed controllers on a 12.7 cm rise, 60 cm tread stair-climbing task, demonstrating our controller improves stability, reduces roll, and lowers noise.