What Are Autonomous Wheelchairs?
As self-driving cars begin to enter the market, it becomes increasingly likely that self-driving wheelchairs will be developed. The implications of this are incredible, and will certainly change the way people roll! This blog series will look at where we are now on the journey towards fully autonomous wheelchairs, as well as some of the pros and cons of self-driving chairs. While I am not an expert on robotics (I’ll leave that to our CEO Pooja), I hope that these insights will help you understand what autonomous technology is and can do!
Self-Driving Wheelchairs: What Are They?
Self-driving vehicles are trickling their way into the market slowly, with Google’s Waymo leading the way towards a 2020 projected launch. The adapted cruise control to maintain distance between 2 vehicles, the lane monitoring software to alert drivers when they are crossing over the line in a road are all already implemented in cars. These technologies make cars safer and easier to drive and are generally considered to be good advances in safety technology. However, trouble arises when you take the human completely out of the equation. Complete reliance on a computer’s ability to make life-or-death decisions properly raises concerns, and the ethics of programming a computer to make those decisions poses issues. Despite this, self-driving cars have been on the streets for a fair while, logging over 10 million test hours (England, 2018), and feeding the AI with data about traffic navigation. This process will take years and millions of dollars to reach the point where you could own a car without a steering wheel or brake pedal.
How it works:
The basic model is that the computer is teaching itself how to drive. By using artificial intelligence (a computer that can teach itself), and inputting millions of hours of driving data into the framework, the computer essentially learns to identify situations. When a car encounters, say a person on the side of the road, it will compare this to the millions of other humans that have been encountered in the past and compute the risk of collision. This will include identifying the probability that the person will step out in front of the car, the speed at which the person is moving, the degree of turning which must occur to avoid the person, the amount of brake that must be applied to avoid hitting them, etc. The car will also need to calculate whether steering around the person will put the driver or other cars at risk, and if so will require a pre-programmed decision-making process to decide whether to swerve, brake or neither. Of course, it is all infinitely more complicated than this, and there are many other factors being considered.
But, we aren’t talking about automobiles, we are talking about wheelchairs, which will likely be more difficult to program to drive safely. Cars operate in fairly controlled environments. On roads, cars and pedestrians observe clear traffic rules (even if they aren’t always followed well- I’m looking at you Toronto drivers!), and although there is some level of unpredictability this is limited. Wheelchairs, on the other hand should be able to travel anywhere someone could walk, meaning the situations that the wheelchair will encounter are pretty much as diverse, unpredictable and lawless as walking through Union Station during rush hour. Additionally, it is likely that self-driving cars will be able to communicate with each other, creating network effects, and helping cars to avoid colliding with each other. People who drive wheelchairs often face challenges with people not getting out of their way, or even walking right into their chair. Communicating with humans is a difficult challenge for autonomous wheelchairs, as warnings would need to be inclusive of people with low vision and/or hearing.
Another challenge will be inputting the desired destination for the wheelchair. While cars can be programmed to travel to a specific address, the input for a wheelchair destination is much more complex due to the large diversity of places a wheelchair can travel.
Take, for example someone at a stadium needs to use the washroom. One possibility is that the person will click a button on their chair that says “bathroom”. The chair will then need to have either a blueprint map of the building, or cameras that can monitor the environment in search of the accessible washroom sign. Using this information, the chair has located the closest washroom! Now, the computer will decide the optimal path towards that washroom. This will require the computer to know the location of all stairways to avoid, and all ramps and elevators (assuming chairs are unable to climb staircases at this point). The path is set, and the chair begins on its way! Dodging people and alerting them to move out of the way, the chair approaches the bathroom. When it approaches, the chair deploys a signal to the door to open, or a mechanical hand to push to automatic door opener. The chair registers that the door is open and is able to move into the washroom!
Once in the bathroom, the chair must be able to choose between the available stalls to locate the accessible one, and the person using the wheelchair may want to back into a specific bathroom stall at a certain angle to make transferring easier. While the wheelchair driver or a human attendant may be able to use their past experience about the easiest transfer method, and therefore best location to park in, a computer may have difficulty accounting for all variables. Assuming this chair has learned from its driver, it successfully docks, and the process must be repeated to return the person to their place in the stadium. The complexity of this decision-making process is high, and potential for mistakes is high as well! A wheelchair colliding with a person is dangerous.
A bathroom is an easy target- what if the driver is hoping to travel to a more specific environment (ie the coffee table to the right of the doorway separating the kitchen and living room). Considering input method must be adaptable for people who have a difficult time speaking or typing the challenge increases. All of these challenges will be faced by developers looking to create self- driving technology for wheelchairs.
While Google Maps and other automobile tracking software has been perfecting available maps of streets and traffic, there are no such maps making blueprints of buildings- this means that autonomous vehicle technology must either find ways of interpreting the environment at a human level of understanding (ie- reading signs, sensing walls and obstacles etc), or every building that self-driving wheelchairs are in must be carefully mapped and categorized.
All autonomous technology is a challenge. It will be years before self-driving cars begin to emerge on the market. As you can see, self-driving wheelchairs pose even greater of a challenge for software developers and thus will likely take even longer to emerge onto the market.
The benefit that self-driving wheelchairs will inevitably bring to the population who uses them is incredible. Working towards an autonomous future for wheelchair controls is certainly a good thing- but the challenges are real as well. Our next post will look at the ethical implications of self driving chairs- the good and the bad!