Snow, Cars and Machine Learning

Photo of Dr. Saha and her husband wearing blue volunteer t-shirts

A Letter from Dr. Ashirbani Saha

Dr. Saha is the first holder of the BRIGHT Run Breast Cancer Learning Health System Chair, a permanent research position established by the BRIGHT Run in partnership with McMaster University.

Hello BRIGHT Run Family,

Hope you had a good start to the new year.

We saw a significant snowstorm in January in Southern Ontario. Hamilton experienced a huge amount of snow. Though it took a lot of effort to manage that amount of snow, I was happy to see children having some outdoor fun.

However, for us adults, driving in the snow is a major concern as we encounter challenges with visibility and control. Incidentally, heavy snow can create a huge problem for fully autonomous vehicles as well.

Before we discuss the reason, let us analyse the term ‘fully autonomous’ closely. There exist six levels of autonomy in driving: Level-0 – full manual control; Level-1– can engage a single automated system in control e.g., steering OR acceleration (adaptive cruise control) with human controls; Level-2 – can engage multiple automated systems such as automatic steering AND acceleration under some circumstances, but human can resume control; Level-3 – vehicles can make informed decisions by monitoring the environment like a human driver, but human can override; Level-4 – all driving tasks are automated within a restricted area (geofencing); Level-5 – can perform like an experienced human driver with just knowing the destination.

A fully autonomous vehicle has Level-5 autonomy.

A car with some sort of automation capability uses sensors such as cameras, lidars (for visibility), wheel rotation sensor (for control). The sensors enable the car to gather information about its own state and the environment surrounding it. Although most modern vehicles employ a decent number of sensors, their usage and accountability are significantly high for Level-5 vehicles when compared to Level-1 vehicles.

Under snowy conditions, the quality of data from sensor(s) is affected (e.g., reduced visibility, slipping like us). The inferior quality of data affects performance (sounds familiar, right?).

Improving the quality or managing with the low quality of information requires an enormous scientific rigor. Thus, researchers and technologists are currently experimenting to improve the performance of vehicles under extreme weather conditions, using techniques for improved sensor fusion (important topic for another eNewsletter!!), machine learning, etc.

That was a heavy dig on automation!!

Let’s get back to digging snow now. Yeah! My husband and I cleared a part of it with help from our kind-hearted neighbour who used a manual snow-blower (something like Level-0). I am interested to know if you would prefer some level of autonomy in your snow removal. What extent of autonomy with this device would make you feel comfortable?