Cybathlon - Legged Mobile Assistance for Quadriplegics

People with severe motor impairments (from e.g. muscular disease, spinal cord injury, cerebral palsy, or neurological conditions) or missing limbs struggle in day-to-day tasks. Currently, they rely on dedicated help from nurses or other caretakers. This limits their independence, which is an important aspect of health and happiness [2]. Automated robotic solutions can provide the necessary tools for people with reduced mobility to perform more tasks on their own [3], and are the future for improved quality of life and healthcare. However, these solutions are still in their early stages [4], and there is a need to both raise awareness and push towards stable platforms that can be certified.

Cybathlon [1, 5] is a competition dedicated to showcasing and advancing technologies that help people with disabilities, e.g., blindness, people in wheelchairs, and prosthetic arms. The Robotic Assistance Race (ROB) in Cybathlon focuses on a robotic assistance platform for people who are quadriplegic. The platform can take the form of either an arm attached to the wheelchair or a completely separate moving robot that the pilot can remotely control. The race consists of a set of everyday tasks that the pilot has to complete on their own, utilizing only the robotic assistant. The tasks in the February Cybathlon Challenge [6] were: collecting a package from a mailbox, brushing the pilot’s teeth, hanging a scarf on a clothesline, and emptying a dishwasher.

Assistive robots will play a key role in people’s lives, as current technologies are being developed for rehabilitation [7], personal assistance [8], and social companionship [9]. Specifically for people with limited mobility who are confined to wheelchairs, most technologies focus on either smart wheelchairs [10] or a robotic arm attached to the wheelchair [11, 12, 13]. With robotic arms directly attached to the wheelchair, the major limitations are reduced mobility of the equipped wheelchair and shared battery supply. In this case, the power and lifetime of the arm have to be limited to prevent the entire wheelchair from becoming inoperational. For example, the commercial iARM [14, 13] arm attachment from Assistive Innovations can only lift 1.5kg [13]. The size is also a limiting factor – too big, and it becomes cumbersome to move with the wheelchair, too small, and the workspace is very limited.

Our proposed solution is the use of an independent quadrupedal robotic platform, ANYmal [15], with a robotic arm, DynaArm [16], on top. This setup offers more versatility than mounting an arm directly on the wheelchair, as it can go places where the wheelchair can not, it has a separate power supply, and it does not limit the wheelchair’s mobility or bloat its size. The independent robot can perform tasks autonomously without the pilot’s direct supervision, e.g. fetch things from another room or empty a dishwasher while the pilot does something else. It also poses less of a safety risk, as any system failures are less likely to affect the wheelchair’s basic mobility, and it does not require the pilot to carry a larger battery and be in constant proximity to a moving robotic arm. To the best of our knowledge, our system is the first walking robotic personal care assistant proposed for people with limited mobility [17, 18].

We showcased our system at the Cybathlon Challenges February 2024, where the pilot had to perform four predefined tasks in a racetrack scenario. The pilot is able to operate the robot through a mouth joystick interface, where they have direct control of its movements. Furthermore, we automate some generic actions, such as bringing an object to the mouth, which is crucial in day-to-day life e.g., for eating or brushing one’s teeth. Finally, to enable unlimited high-level instruction without the need for tedious GUI menus, we integrate a voice command interface. Our system won in its category in the challenge, achieving a full score and a best time of $6$ minutes $34$ seconds. Although some of the task descriptions in the challenge are very specific, in this work, we present a versatile general-purpose system that can perform a variety of other tasks and actions.

The system consists of three main components: (i) an ANYbotics ANYmal D body is the base of the robot, allowing for quick and robust locomotion across the track; (ii) a Dyna-Tech DynaArm, attached to the top side of the body and used for precise manipulation; (iii) a Quadstick gaming controller which is used as the main control input device by the pilot. Our system is operated by Samuel Kunz, a mechanical engineer who became quadriplegic after a swimming accident in 2014. An overview of the system and the main components is presented in Figure 1.

As the system is mainly teleoperated by the pilot, their proficiency is integral to its performance. How much training time is needed depends largely on how intuitive the control scheme of the system is [14]. During training, the pilot practiced executing the challenge tasks on a mock race track and gave feedback for the refinement of the control scheme and GUI visualization of the system. Our pilot, Samuel Kunz, trained for roughly 15 hours over five sessions to reach the level of proficiency exhibited during the February Challenge. This included minimal failure rate executing the race tasks, collision-free movement of the robotic base across the race track, and synchronized movement of the robotic base and the motorized wheelchair.

The February Challenge consisted of two runs, and for each team, the run with the best score (best time on equal scores between runs) was counted as the final result. The track setup is presented in Figure 4. We achieved a full score on both runs with times of 6 minutes 34 seconds and 6 minutes 51 seconds. Additionally, we were the only team to achieve a full final score. Throughout our runs, we spent on average, 23 percent of the time moving along the track and 77 percent of the time manipulating. The motion on the track includes both the robot and the pilot moving into position. While most of the time is spent on manipulation, automating many of these tasks is difficult, as making perception pipelines as reactive as direct human control is still an active challenge in the robotic community.

Although Cybathlon is a fairly pre-defined competition scenario, our aim was to develop a generic control method that also generalizes to previously unseen tasks. The success of this can be seen in our performance on the Scarf task, a task which the pilot had not trained for during practice on the mock course and nonetheless completed with a full score and a best time of 64 seconds.

We have presented the first design for controlling and operating an independent quadrupedal assistance robot for people with reduced mobility (e.g., quadriplegic). Our platform is highly versatile, as it can reach places that remain inaccessible by wheelchair and does not force the pilot to permanently carry a robotic arm attachment. This design won the Cybathlon Challenge February 2024 Assistant Robot Race, where it completed a series of everyday tasks in record time. In the future, when assistance robots are medically certified and can perform more tasks autonomously, they will have a huge impact on the lives of many people who need daily care.