Month: August 2021

Do Robots get the “Twisties”?

In watching some of the Olympic events and the Paralympic events, you can’t help but be amazed at the human body.  There has been a lot of discussion about Simone Biles experiencing the “twisties” and how that  danger puts an athlete in danger..  While my own experience as a gymnast didn’t get much past cartwheels, I am acutely aware of what this phenomena can do.  Humans utilize proprioception, which is the awareness of the position of the body, to control our bodies.  When that awareness is altered, we can have trouble with a host of things such as reaching, balance, and walking (not to mention executing multiple aerial flips).  Robots also have a concept of proprioception through sensors and calculations.  Exoskeletons aren’t doing backflips (though you can see some impressive gymnastics from Boston Dynamics’ Atlas here:, but accuracy and control of the motion they are executing is still important.  When I design the gait trajectory of an exoskeleton, the placement of the foot in space throughout the gait is very important.  If it is too high and the human cannot match the awkward high-step gait; too low and the human will trip when their foot does not clear the ground.   Toe clearance can be as low as about ½”, so there is not a lot of room for mistakes.

If we want to know where the exo’s toe is, there are a number of ways to do this.  One is to directly measure the distance from the ground.  To do this while walking, you need something like a laser range sensor or sonar sensor.  These can be difficult to ensure that your measurements are accurate on a variety of surfaces. 

You can also use IMUs (inertial measurement units).  These utilize accelerometers and gyroscopes to calculate orientation and acceleration and these measurements can be integrated to determine position.  These sensors are subject to drift, though there is extensive research on how to mitigate this problem which I will not cover here (but it’s worth a Google search if you’re interested!).

One of the most common ways to measure toe location uses forward kinematics.  Forward kinematics takes the known position of one link of an object and calculates the end position using the known lengths and angles of each.  

Take for example, a 2-link robot arm.  The base is fixed on the ground, so we’ll call that (0,0).   Each of the links have an angle and a length associated with them.  If we want to calculate the position of the end of the robot, we use trigonometry to calculate the X and Y component of each link and then add them together.

A 2-link arm

    \[x = l_1cos(\theta_1) + l_2cos(\theta_1+\theta_2)\]

    \[y = l_1sin(\theta_1) + l_2sin(\theta_1 + \theta_2)\]

But what if the robot has the “twisties” and the measurement of the first angle is off by just a little bit?  Maybe the sensor calibration wasn’t quite right or the mechanism got loose and slipped a bit or the beam isn’t perfectly stiff and so it sags a bit through the length?  Let’s a assume an error of just 5°.  It may not look like much to our eye, but it can mean the difference between successfully picking up an object or colliding with it.

Let’s put some numbers in just to see what difference a mere 5 degrees makes. Let:

    \[l_1 = 12 in\]

    \[l_2 = 10 in\]

    \[\theta_1 = 30^{\circ} \]

    \[\theta_2 = 15^{\circ} \]

In our original calculation, y = 13.07 inches.

However, if sensor is off by 5°, we have to adjust the measured angles…

    \[\theta_1 = 25^{\circ}\]

    \[\theta_2 = 15^{\circ}\]

Now y = 11.5 in.

When we look at the kinematics of a body, we usually start from a foot that we know is on the ground and work our way up that leg and down the other to determine the foot position.  Over this many links, small errors in our perception can cause big problems.  5° is as simple as not measuring the flexibility of a link or accumulated small errors due to the inaccuracy of sensors.  Given this, I am constantly amazed by our bodies and our ability to do things as seemingly simple as walk across a room to as complex as flips on the balance beam.  We engineers have quite a challenge ahead of us!

Fitting an Exoskeleton

Exoskeletons derive their name from the exoskeletons we find in nature.  Those seen on insects fit precisely along the body, enhancing the strength of the insect while allowing for motion.  The walking stick shown here gains strength from this supportive structure, but the joints are intricately developed to work with the muscles and tendons to allow for a very large range of motion at the joint.

photos: ©Matthew Strausser IG: @strausserphotography

A rehab exoskeleton must likewise become an extension of the patient, working with both their own muscles and the actuation of the device.  Therefore, the joint alignment and sizing is critical to comfort and functionality.  But our challenge of time remains and we must consider the tradeoffs between accurate sizing and speed.  If we adjust every aspect of the device perfectly, the device will fit snugly, but each patient will either need custom pieces or set up times will become prohibitive.  If we do not have sufficient resolution, however, we risk pressure sores, abrasion, discomfort, or muscle strain.  We as engineers are thus challenged to develop adjustments that are fast, accurate, and secure and use compliance where rigid attachment may not be necessary.

EksoNR utilizes adjustments at the upper and lower leg and hip width as well as velcro soft goods to achieve a safe and comfortable fit while being relatively quick to set up. 

Next, the patient must get into the device.  For some devices, the patient can put on pieces while sitting or may need to put an AFO or similar brace into their shoe.  In others, they transfer into the device, and still for others a bodyweight support system may be used to help get that patient in. Locking in the adjustments may require specialized tools, and we must consider their maintenance as well.  It is advantageous to have adjustments that an untrained user, such as a therapy aid, can make as this decreases the time out of the therapists’ session.  We also consider what adjustments may need to be made during the session and how to accomplish those with minimal disturbance.  We also have to provide the therapist a way to determine if the adjustments are correct whether through visual clues, messages from the device itself, or tests that can be run to check.

Currently, most devices use traditional mechanisms for locking and positioning, such as nuts and bolts, friction clamps or pins.  Improved materials and manufacturing processes will provide us new opportunities for improving and streamlining adjustments. For example, printing carbon fiber or even metal components allows us to gain strength while using intricate patterns for tool-less locking.  There are also soft sensors and soft actuators in development that will improve fit and self-adjust support and bracing.  The challenge here becomes balancing the complexity and cost with the benefits gained in terms of therapy outcomes.  As these technologies advance, the balance may come to tip in the favor of additional sensing and robotic controlled adjustment, but I will leave that for a future design project.

The Legacy

The Legacy

I was introduced to exoskeletons as a graduate student at the University of California, Berkeley where I was fortunate enough to study with Professor Kazerooni whose work focuses on enhancing human capability through exoskeletons.  The lab developed devices like BLEEX, which give able-bodied individuals added strength by wrapping a mechanical structure around the body and adding power.  

The initial devices were large and required offboard power or loud engines which limited their use.  Then the lab developed HULC, the Human Universal Load Carrier, which was a breakthrough exoskeleton developed to help soldiers carry load through unknown terrain. 

 It followed the same model of BLEEX and other exos in mimicking and adding to the body’s structure, but it utilized gravity and joint alignment and actuation in a way that enabled the device to be run off of on-board power.  HULC’s design transfers the load weight around the human body to the ground and added power to the legs only where needed. Professor Kazerooni saw that this technology could be extended to add power where there was a deficit, not just where excess power was needed.  My colleagues and I began work in developing the concept of a medical exoskeleton and were met with many challenges- some expected, some not.  I was excited by this challenge and the potential to help those who had suffered from a stroke or spinal cord injury, and while I knew the challenge would not be an easy one, it took some time to understand the difficulty of transitioning from developing a device for able-bodied load-carriage applications to one that provides mechanisms for gait rehabilitation.  

An exoskeleton for load carriage does not need to provide assistance in all directions.  Gravity helps with knee flexion (like when squatting down), so power can be limited to extension.  In a rehabilitation device, the patient lacks power or control in both flexion and extension so both must be powered.

In a device built for rugged terrain, the soldier must have full flexibility.  Therefore, joints that are not controlled must be free, such as hip abduction to allow for side-stepping.  In therapy, we must control all joints even at the cost of reduced freedom by locking them out completely.  

HULC was built for use by the army, to be used outside and with trained soldiers.  Service could be completed by the users, so regular maintenance was not an issue therefore allowing us to utilize hydraulics to optimize the location of components.  In the medical setting, however, regular maintenance and potential leaks were not acceptable while the distribution of weight was less critical, so a more traditional motor system was chosen.  

A soldier must be able to quickly take off the exoskeleton for full maneuverability in case of emergency or attack, thus the robot has minimal connection points.  In a clinical setting, the patient must be tightly coupled to the device for safety and while donning and doffing must be fast, it does not require the speed a military device does.  However, the user must be able to transfer into it with limited to no motion in the legs.

There are many additional challenges that face rehabilitation exoskeletons.  While some are similar to those faced in the military or industrial applications, many are unique to the users and settings found in rehab.  When I entered that lab more than a decade ago, I did not realize how much this problem would draw me in and become my passion.  The challenge of building a robot for gait rehab is one that challenges the engineering mind while never letting us forget who we create for.