Why make a walker that can trip people?

In rehabilitation there has been an increase in the use of robotics for research purposes. This often takes the form of software written for commercial robots. The design of hardware systems for rehabilitation in robotics provides another avenue for studying human behavior. For assistive device users, especially, technology often plays a large role in their day-to-day life by reducing the effort for daily activities like walking.

To study real-world behavior in assistive device users, we designed a piece of hardware that assistive device users recognized and combined that with robotic actuation to generate new experimental conditions. While designing a prototype, there are guiding criteria the prototype must meet. For us, this was to create a research tool to provide overground perturbations, or small disturbances to balance while someone is walking on a flat surface other than a treadmill. We wanted this research tool to be used in experiments to collect data in tandem with a motion capture system, to see how people recovered their balance after being perturbed.

Balance studies are often conducted with people post-stroke, 26.1% of whom use walkers. Although current perturbation techniques can be adjusted for assistive device users, there are unique risks that come from perturbations during walker use that are not well understood. Specifically, there is an risk of falls, which occur mainly during walking or from a trip or slip. This could be due to the physical characteristics of the walkers, such as their weight and bulkiness, which may result in perturbations from the user colliding with the environment or the walker itself.

Mechanical Design

Walker choice

In order for an assistive device user to be able to use our device in experiments, we chose to keep our device as close to a physical walker as possible in design. We chose to modify a rollator because it is the most popular type of walker and use of a rollator is associated with risk for injury due to falls. The Drive Rollator (Drive Medical) we modified has four 6.75” wheels and a basket. In addition, we added adjustable rods inside the handles to extend the height for taller users. We named the modified rollator FALL-E, inspired by its purpose to trip people.

Motorized back wheels

The main criteria of our design was to create overground perturbations. To do so, we took inspiration from the method used by treadmill perturbations, where the belt speed of the treadmill is abruptly changed. We first recorded acceleration of a walker while bumping into various objects and rolling over different floor terrains to simulate the perturbations we wanted to mimic. This gave us the range of accelerations in typical perturbations (1-3 m/s²).

With our accelerations and after several rounds (and 3 years!) of design revisions, we settled on replacing the back wheels with two hub motors with Hall effect sensors (Parallax, SKU #27860). The hub motors came with attachments that we designed around to make blocks to attach to the walker legs through the existing axel pin. We made modifications for the wheel height mismatch with the hub motors and the Drive rollator wheels. Image B shows all these components combined with the load cell housings to form the back wheel assembly.

Load cells

The load cells in the legs (outlined in yellow in image B above) are used to quantify the amount of vertical load users applied to the device. This was due to an experimental design criteria; vertical load can be used as a proxy measure for upper body stability. We used four 50 kg button load cells (Phidgets, ID #3136_0), one in each rollator leg, for their low cost and ability to distribute load.

The machine shop made the load cell housings by cutting the rollator legs and installing them in-line. These housings are mounted to each wheel and consist of two parts. The top part is a machined rod mounted to the leg and slides inside the bottom part, which consists of a linear bearing tightly fit inside an outer tube (with grease) with the load cell fastened at the bottom above the wheel. For calibrating the load cells each bottom part with the linear bearing, load cell and wheel, can be removed and calibrated off the rollator. The example calibration setup is shown below.

Electronics and Software

Now that we have FALL-E, how can we make it come alive and move in an experiment? We need it to perturb in a controlled way as well as record and sync with other systems we are using in the experiment. Electronics and software will help us do this. Apart from the hub motors with encoders, the load cells and a switch on the handle for motor state control, all of the electronics are housed in FALL-E’s basket in a three layer design, shown below.

The bottom layer holds the 5S Lipo battery (GetFPV) in a fireproof bag and a voltage regulator between the battery and breakout board (Digikey).

The middle layer is the busiest, with a breakout board for the Teensy 4.1 microcontroller (PJRC) that helps with voltage regulation for the Teensy and provides mounting for other circuit components (Protosupplies), analog-to-digital converters for the load cells (Sparkfun) and an audio shield (PJRC).

The top layer is mounted upside down to the basket lid, to keep electrical components as far apart while still fitting within the basket size constraints. This is where the two ODrive S1 motor controllers and brake resistors that control the DC brushless hub motors sit, as well as the buttons to control the rollator.

Control

Most of the code is written in C++ and run through Arduino for compatibility with the Teensy 4.1 microcontroller. Motor speed is controlled with a proportional (P), integral (I) controller as well as real-time adjustments made to the feedback gains based on the measured load applied to the rollator. In effect, this is PI control with gain scheduling based on the load to consider how added weight affects the acceleration of the rollator determined by Equations 1-3:

The data logging software uses custom and modified libraries to record time-stamped data in real time from the motors, load cells, buttons, and states of the software. Dynamic memory is used to create packets that are saved to a local micro-SD card on the microcontroller when the memory buffer is filled. Data is logged at 100 Hz.

FALL-E is controlled using a state machine to transition between idle and experimental states (pictured below). Idle mode allows the motors to be backdriven with no resistance, as if they were just regular wheels. We added a switch on the right handle to set the motors in idle mode if there is a problem with the perturbations.

Links for additional technical details

Github with repo for FALL-E: https://github.com/cyunis/FALL-E_public

STEP file of the CAD model for the wheel mount blocks: https://drive.google.com/drive/folders/1w6PVUXSH3zrylVI88UIOxVl8JBmlHrDK?usp=sharing

FALL-E in Use

When we put the mechanical design, electronics and control software together we can see how FALL-E performs. FALL-E is designed to provide overground perturbations randomly while someone is walking, simulating a situation where someone accidentally bumps into something or rolls over an obstacle or rough surface with a rollator. These overground perturbations were designed to be performed during an experiment in a motion-capture lab. The lab we used had a 9 m long walking space, 6 m of which was in view of the motion capture cameras (Qualisys). FALL-E and the research participant wore reflective markers to track the position of anatomical landmarks on the body and key points on the rollator to collect kinematics at 100 Hz. We used a sync cable connected to both FALL-E and Qualisys that recorded experiment onset simultaneously with a sync signal provided by button presses.

To align the datasets from Qualisys and FALL-E in time, we take the offset between the sync signals and apply it across all timestamps from FALL-E. This is because when FALL-E’s program loads off the computer at the beginning of the experiment there is sometimes a lag during compilation that creates a time offset from the computer clock on the Teensy clock. Once the timestamps have the offset applied, we can then plot data from both sources together, as in the graphs below.

People use rollators with different postures and lean onto the rollator with different loads. To check the ability of FALL-E to accelerate and decelerate to a given speed with and without load, we automated the motor speed to 1 m/s and then applied a perturbation with the walker loaded with varying weight. Perturbations were either a 50% acceleration, 50% deceleration or 100% deceleration in speed collected 5 times per condition over a 0.5 second window.

The image above shows a typical trial during which a participant walks across the lab and a perturbation occurs at some point while they are walking. The graph shows how the actual speed of FALL-E may be influenced by the participant, and may cause the wheels to not reach the final set point speed due to wheel slip or additional resistance torque applied by the participant on the wheels.

Data Analysis with FALL-E

Since FALL-E is designed to be used for research purposes, several variables are logged locally during use. These include: the setpoint speed, the PID input for right and left wheels, the right and left encoder speeds, the four individual load cell weights, the two button Boolean values, the number of perturbations that have occurred, and the perturbation condition being set.

Future implications

Robotics have played a role in daily life through assistive technologies such as hearing aids and socially assistive robots. Using robotics to create unique test conditions to study mobility in disabled users can provide opportunities for the study of minority populations, as well as create interesting new ways to modify assistive devices that people already use every day. One of the most common things people say when they demo FALL-E is how excited they are for a motorized walker, and how helpful it could be. Maybe with the rise of e-bikes, motorized scooters, and other motorized mobility aids, motorized walkers can not only trip someone for a balance experiment, but also assist people to go where they want to a little more easily.

Acknowledgments

Thank you to the Musculoskeletal Biomechanics Research Lab for the experimental space, Dornsife Machine Shop for designing and creating the load cell housings for the legs as well as machining the blocks to mount the motors to the legs. Thank you to Patrick Campbell for code modified for data logging purposes. Thank you to Marc-Daniel Julien for code review.