Data Analysis — Beta Test Ride, Chennai
During the last weekend of August 2016, we hosted four Open House sessions at IIT Madras Research Park in Chennai. Considering our roots in that very campus, it was like homecoming for us. It gave us a chance to interact with a lot of enthusiasts, some of whom have been following us since the days when we were a small team still curiously attempting to build our first battery pack.
Besides sharing our vision towards a future of smart connected rides during these sessions, we gave a chance to some folks to test ride a beta version of the S340 for them to get a hang of the vehicle.
During their test rides, we collected the data from our enthusiasts about their ride; equipped with sensors to capture different dimensions of one’s ride, our smart scooter logs multifaceted data about the ride — from scooter orientation and GPS to the electrical and thermal state of different components.
Post their test ride, we used this data to present a summary of their ride analytics — from basic ride statistics to data on ride intensity and efficiency.
The Test Ride Track
With the weather gods plotting a heavy downpour in the Research Park campus on that day, we had to conduct the test rides in a rain proof area. So we chose Basement 2 of IITM Research Park to be the ride track, which turned out to be incredible considering the covert nature of the location and how that gave the vibe of exclusivity to our enthusiasts.
We marked out the ride track, approximately 210m long, using traffic cones to convey ride path and warning tapes to notify the rider about turns and change of direction.
Mapping a ride without GPS
One drawback of conducting the test rides underground was that the cell phone and GPS signals were spotty, which meant that we would have no data on GPS coordinates to perform ride analyses. But we sensed an opportunity to attempt estimating the track coordinates using Inertial Measurement Unit (IMU) data in the absence of GPS.
The IMU measures state variables of motion such as acceleration and angle of inclination. Using this data, we calculated the direction of the vehicle at each point in the ride and consequently mapped out the path taken by the rider, as shown in Figure 1 below.
Having demonstrated this capability thus, we believe it’ll be extremely beneficial in use cases such as riding through a tunnel or near tall buildings and trees which obstruct GPS signals. While even popular GPS based apps would need external assistance or simply yield with a ‘No Signal’ message, our vehicle would continue to seamlessly navigate!
Detecting an erratic ride
In this section, we will discuss how our vehicle intelligence can detect erratic riding patterns that are susceptible to mishaps.
But before that, let us take one of the test rides from that day and observe its dynamics.
The rider accelerates when there is a long straight stretch ahead, and slows down near the corners, and for this reason speed is highest in the middle of the straight stretches, as visualized in the Figures 2 and 3 below.
Note: All the numbers shown in the charts of this blog post have been rescaled proportionally based on actual values for visualization purposes, they do not represent actual values. Negative acceleration values indicate deceleration.
Next, let’s look at how the g-forces experienced by the rider change during his ride, through two state variables — 1) longitudinal intensity, which is the force that one experiences while braking (lies in the direction of motion of the vehicle) and 2) lateral intensity, which is the force that one experiences while leaning to make a turn (normal to the direction of motion of the vehicle).
From Figures 4 and 5, we can observe that the longitudinal intensity is mostly high at the onset and end of a straight stretch, while lateral intensity is high at the corners, as expected. The intensity values are positive or negative based on the direction of the force.
Now comes the interesting part. We visualized the g-forces experienced at each moment of all test rides in a single g-g plot shown in Figure 6.
While most points form a cluster at the center of the chart, we observe a set of points that spill out of the envelope — they have highly negative longitudinal intensity for given lateral intensity. What’s noteworthy about this set of points is that they are all from the same ride — one during which the rider lost control of the scooter and had a fall.
We observe a pattern here — that a ride is regular (i.e. does not result in a mishap) when the lateral and longitudinal forces experienced by the rider are enclosed in an elliptical envelope as illustrated in Figure 6. When the locus of g-forces of a particular ride falls outside of this region partially or wholly, the ride is erratic and likely to result in mishap.
So what can we do with this understanding? For example, we can notify riders who are constantly hovering near the edge of these envelopes that they are at a risk of suffering a fall, and provide real time warning to avoid mishaps from occurring.
Also, in such events where the rider falls off the scooter, we can detect the exact spot of the mishap by tracking the rate of change of the lean (or roll) of the scooter as seen in Figure 8. There is a sudden spike in its value where the fall occurs. Once we detect such an event, we can offer options for help via the dashboard based on the possible severity of the situation.
Another interesting thing to see in this dataset is the change in the rider’s behavior post their fall on the track. Plotting the magnitude of the overall forces experienced by the rider on the track (Figure 10 below), we can see where the user falls and decides to take it easy from then on.
The Pareto Frontier
Have you wondered about how well you’re riding your vehicle? With vehicular data, it’s possible to analyze the overall rider performance and efficiency.
Let’s look at a plot of energy from the battery consumed during the ride versus time taken to complete the ride by the rider (Figure 11). We can observe that in general, when a high amount of energy has been used for the ride, the lap times tend to be lesser.
Using too much energy or too much time to complete the ride is inefficient. Therefore, ideally, one ought to be at the bottom-left corner of the graph, close to or below the high efficiency curve. You see most of the riders that day tried to get close to a boundary around the optimal, which we call the “Pareto Frontier”. Beyond this curve, if you are lowering your energy consumption, you are compromising on time, and vice-versa.
We will make the ride data available to the riders, so they have a chance to see how their riding behaviour is. Using this information, riders can choose to improve on their habits which make them an inefficient rider, and hence increasing the range and life of the S340.
There are a lot of possibilities to explore with the ride data that is available to us. We hope this blog illustrates a part of how we are using it to enhance the rider experience — from better navigation, to giving feedback on the user’s riding style to detecting mishaps and distress management.
A lot of these services get better over time, as we get more data to analyze and find patterns. So you can expect the benefits you get from it to get better every day.
We are currently gathering data to build these services using test rides such as this, and by equipping existing scooters with our ‘Make Every Scooter Smart and Intelligent’ (MESSI) systems, so that we can develop these services in parallel to the development and production of the S340.
We will keep you updated on further developments on this front, so keep following us!
Originally published at www.atherenergy.com on October 7, 2016.