Can Quadruped Guide Robots be Used as Guide Dogs?

RQ1: Functionally, how many differences in guiding experience between quadruped robots and wheeled robots, and what are the potential causes?

RQ2: Emotionally, what kind of emotional experience can the bionic forms of quadruped robots bring in guiding?

RQ3: What are the special insights of designing a quadruped guide robot from the perspective of BLV users?

23 participants (7 females, 16 males, 28.870 $\pm$ 6.608 years old) were invited to participate in the indoor and outdoor walking experiment (14 indoor participants, nine outdoor participants), including eight sighted participants, expected to simulate the state of suddenly acquired blindness [19], and 15 blind participants. We collected basic information about the participants through a questionnaire. We screened them according to two criteria: ignorance of the experimental route and the capabilities of the two forms of robots (such as operation methods, etc.) (TABLE I).

Quadruped Robots: The quadruped robots used in the experiment are both in bionic forms (Fig. 2), which is a mechanical structure with 12 degrees of freedom realized by 12 high-performance servo motors. They both inherit the open source force dynamic equilibrium of MIT Cheetah [20, 21], with good multi-terrain adaptability. Basically, they can realize the necessary movements in navigation, such as forward and backward, left and right movement, turning in place, up and down slopes, etc. Leg movements are limited to various gaits. The indoor experiments used the Cyberdog from Xiaomi, 771×355×400 mm (L×B×H), while the outdoor experiments used the A1 from Unitree, 650×310×600 mm (L×B×H).

Additionally, guide dog users routinely use a guide harness, which is divided into a harness (adapted to guide dogs’ bust size of 640 to 940 mm) and a tie rod (450 mm), to perceive the guidance of the guide dog better. Therefore, the quadruped robots in this experiment are equipped with an entire guide harness like guide dogs.

Wheeled Robots: As the comparison group for the quadruped robots, the size of the wheeled robot used in the indoor experiment is designed to be the same as Cyberdog, 771×355×400 mm (L×B×H), and the outdoor experiment selected the Guiding Car [22] with the size of a guide dog as a reference to better accommodate the outdoor complicated environment. In addition, the two-wheeled robots are respectively equipped with a tie rod of guide harness and an adjustable handrail by rigid connection to achieve a similar guide experience to quadruped robots.

Control & Operation: Considering the complicated environment and technology limitations, we unified the remote control system to realize the Wizard-of-Oz method. The controller was a 360° rocker, transmitting data such as moving direction, speed, and emergency brakes to the robot wirelessly. It was tested that the operation delay of the remote control system was less than 100 ms, enough to meet the needs. During the indoor experiment, the operator stood in the non-test area and operated the robot to follow the ground route. In the outdoor experiment, the operator was always located 3m behind the participant and stopped operating for a short time in case of an emergency. To eliminate the influence of the operator’s subjective factors, in each experiment, all instructions to two kinds of robots were made by one operator, and she accepted operational training before the experiment to ensure robots guide participants fluently.

Indoor environment: We carved out an indoor experimental area of 8×8 m in the center of a 9×10 m lecture hall, with a 6.5×6.5 m area as the walking task area (Fig. 3). The base station of HTC VIVE was installed in the four corners of the experimental area (8×8 m), 2m from the ground, to cover the entire area. Then, by tying two Trackers to the participant’s waist and the robot’s body, the Tracker could receive the infrared light signal emitted by the base station and give feedback. Through HTC Vive Tracker, we can obtain real-time position coordinates between participants and robots.

Outdoor environment: To explore the performance of two forms of guide robots in real daily life, we asked the participants for some desired destinations that they have difficulties reaching alone nearby their neighborhood in advance. And two routes were planned first. After trying to simulate the real travel conditions of the BLV user, we then made some adjustments to ensure that the two walking routes both include brick surface, uneven brick surface, tarmac, narrow road, obstacles, and uphill/downhill. These grounding conditions include the majority of areas where the BLV often visits. We did these to allow the BLV to experience as many different chassis of robots interacting with different road conditions as possible while ensuring their safety and preventing fatigue, and thus to evaluate different chassis of robots fully. We also considered stairs, but considering the limitations of current quadruped and wheeled robots, we gave up the plan of navigating the stairs. The total length of the two routes is about 100 meters. The detailed grounding information and environment can be seen in Fig. 4.

Firstly, all participants were invited to know the detail of a consent form and sign up for it. Because of the difference in visually impaired degrees, which means some participants have feelings of light or see the shadow of stuff, an eye patch was worn before the experiment. We also paid attention to the disruption of sound from robots, which were controlled by wearing earphones to hear the same sound. But in the outdoor experiment, to simulate a real outdoor walking scene and prevent risks, we didn’t ask participants to wear eye-patch and earphones anymore. Then, due to challenges related to using the guiding harness, we set up a session to train the use of it.

After preparation, participants were invited to finish walking tasks. We designed an indoor study to have four trials for each chassis. Trials combining different tasks and difficulties were presented for each participant by reverse balance method to decrease sequential effects. For the difficulty, we used easy and difficult settings. And for the task, we used the straight & Turn (S&T) and Obstacle Avoidance (OA) settings. About the definition of the difficulty of tasks, we take rotation angle and the number of obstacles into consideration. For easy S&T tasks, we arrange the robot to lead the participants to make three 90-degree turns. For difficult S&T tasks, robots were set to turn at different angles(45 degrees, 90 degrees, and 135 degrees) [23, 24, 22]. Two identical chairs heading in different directions were placed in the laboratory for easy OA tasks. For difficult OA tasks, three identical chairs were placed, and participants had to avoid the obstacles more frequently in a short distance, and the arc length of avoidance was longer. (See Fig. 5)

During the outdoor experiment, they also went straight and turned, including right angle turn, and larger and smaller angles. Avoiding obstacles also happens naturally, such as pedestrians, bicycles and wire poles.

After each experiment, the participants were invited to rate each robot’s usability, workload, satisfaction, and trust on relevant scales (details are in part F). Finally, the participants were invited to accept a semi-structured interview about their general experience with robots. Such questions as ”what do you think of the advantages of quadruped and wheeled robots?” and ”how do you feel when you turn or avoid obstacles with quadruped robot and wheeled robot?”

About the questionnaire, the subjective valuables were categorized into the below two aspects: subjective assessment of robots’ use (usability, workload) and personal feeling (trust and satisfaction). All questionnaires have passed the reliability test, and their Cronbach’s alpha is acceptable.

1. 1.

   Workload (NASA-TLX): The National Aeronautics and Space Administration Task Load Index (NASA-TLX) scales were used for assessing the ease of use[25]. This scale includes six items: mental demand, physical demand, temporal demand, performance, effort, and frustration. All participants were asked to rate from 0 to 20.

2. 2.

   Usability (SUS): In 1996, the System Usability Scale (SUS) was created by Brooke et al. [26]. It assessed usability via a five-point Likert scale ranging from 1, completely disagree, to 5, completely agree, replacing all the words ”system” with ”robot”.

3. 3.

   Trust: The questionnaire was adapted from the seven-point trust in automation scale by Jian et al. [27].

4. 4.

   Satisfaction (QUEST): The Quebec User Evaluation of Satisfaction with Assistive Technology (QUEST) Scale by Demers et al. [28] measured the satisfaction level. 1-5 scores were rated as the level of satisfaction.

Previous studies showed that older people preferred realistic, familiar, bionic designs over mechanomorphic, and such designs increased the incidence of emotional responses and interactions[17]. However, we did not know how robots in dog form would affect the BLV users. Thus, we studied whether the group could also empathize with quadruped robots. One observation was that most participants can neither feel quadruped robots were friendly nor could bring them a range of positive emotions, such as companionship and cordial feeling. Notably, one of the themes repeatedly emphasized was practicality, like P10 said, ”The appearance is not important as long as it’s portable. It is the guiding capability that matters.” According to the description from guide dog users, despite their passion for guide dogs, they have no preference for the bionic form. Whereas, there were still two participants( P17 & P21) who expressed that the robotic dog was affectionate, which they described as if they were walking a dog, yet they were both sighted people. Similarly, previous research had organized focus groups to learn about BLV people. When the participants expressed their perception of a guide robot, some BLV people imagined that they might want a dog-like robot [37]. However, when the BLV people in this experiment used the dog-like robot, they realized safety and comfort were the most important guiding aspect.

The above finding revealed the potential improvement direction for quadruped guide robots. The underlying design logic of quadruped robots gives them the ability to make greater breakthroughs in finishing complex tasks, such tasks as getting on the steps[38], trotting over obstacles[39], dynamic grasping[40], and so on. Therefore, we hope the following insights could be incorporated into the future iteration of quadruped guide robots to improve the walking experience while maintaining functionality.

Developing harmonic and gentle moving gaits could decrease unnecessary feedback and enhance the sense of certainty. Naturally, anxiety from being stepped on could relieve it to a larger extent. However, a meaningful topic is what kind of and how much information is appropriate for BLV users, and will various feedback lead to being confused or feeling safe? Moreover, adding professional actions of guide dogs, such as stopping on the front paws before going up the steps, will also greatly help the BLV users.

Besides, participants repeatedly mentioned the noise problem inspired us to consider broad possible occasions of use as BLV users sometimes share similar social life with most sighted people. Quadruped robot does well in complex terrain conditions, but social life is similarly complex to think over. The application of irregular and variable shapes or flexible materials can be one solution to reduce the walking noise of quadruped robots.

Simultaneously, the findings on BLV users’ emotions toward quadruped guide robots give an inspiration that even though the combination of ”dog-like” and ”walking” is associated with the implementation of quadruped robots as guide dogs, much efforts on the more realistic biological-structure simulation is still needed to give BLV users a similar sense of trust and satisfaction as biological dogs.

Other robot chassis can be considered in the future, such as wheel-legged robots, which compensate for the weaknesses of each independent chassis. Now wheel-legged robot combines the strengths of both chassis like it will move in rough terrain and roll in a flat surface[41]. [42] developed Wheel-Track-Leg Hybrid Mobile Robot with great obstacles-crossing capability, and [43] developed a control system to customize the tail according to obstacle parameters. These breakthroughs are all beneficial for solving problems during guiding. Therefore, it would be a great choice to provide better guiding performance and experience with the help of wheel-legged robots.