The Grid > Eloy HQ > Zenzic CAM scale-up: day 2 planning and debrief

Zenzic CAM scale-up: day 2 planning and debrief

With the break over Christmas and the New Year, we didn’t have time to write an update on planning for our second day of testing at UTAC on the 6th January so we’ve combined it with the debrief here.

Our Day 1 of testing in December went well but we had a lot of open questions about how to measure efficacy – perhaps too much of a goal for a first day but an area we hoped to be clearer on after Day 2. However with robust data collection and some tweaks to our Artificial Intelligence MVC, we were confident of being able to assess where value and benefit can be created with these types of driver interventions. Let’s discuss what happened!

Planning

We decided to follow the same 2-session structure, with 2 hours of testing in the morning and 2 hours in the afternoon. Luckily, we had the same 10 drivers, which gave us substantially more time to improve the track setup. This time, we used 60 cones as well as red and white tape to close off one lane in the narrow stretches of road so that cars couldn’t pass each other as they had on Day 1. We also reduced the number of narrow sections from 5 to 4, with the “wiggly section” removed because we determined the curvature of the road was too much of an edge case for this stage of testing.

Weather

Weather had been highly variable in the previous month, with minus 8 Celsius and snow before Christmas, and heavy rainfall during and after. Luckily, the 6th was relatively warm and sunny which meant standing outside for long periods was bearable and the drivers were able to see the track clearly.

Our co-founder Anna, setting the cars off on their laps

Filming

As part of the Zenzic CAM Scale-up programme, a film crew was also present for the day. We will have far more details about this in February, hopefully with some excellent footage from the day.

Session 1: app on (driver intervention)

In the morning, we wanted to test the software with the driver alerts switched on. It was easier to get this all ready for the start of the day and then run more base case testing with no app running and no interventions in the afternoon. With the film crew present, this meant we could focus on key software testing and technical improvements during the first session and use some of the afternoon for rerunning scenarios that might have been missed from filming.

Session 2: app off (baseline)

The afternoon continued to be dry (and sunny). With the MVC turned off, the phones were used to track the vehicles as they moved around the course. As discussed on the Day 1 debrief, with relatively low vehicle density and a good line of sight, human drivers are relatively good at navigating short, narrow stretches of road. Having 10 professional and experienced drivers in mid-sized and small vehicles also helped. The traffic in the 2 inner, narrow sections tended to move well with and without the software running.

However, the other 2 narrow stretches had less visibility, so when making a decision to proceed through the narrow section, the drivers were unable to see if another vehicle was approaching from the other end. Due to this, every time 2 vehicles entered from opposite ends they met and 1 vehicle needed to reverse. As subsequent cars entered the narrow stretch, a small traffic jam formed, as determining who should reverse became harder for drivers to communicate across several cars.

Efficacy data

Much of the data we collected will be useful proprietary information but we have some initial findings which I will share here. Average lap time is a crude method to assess quite a complex set of dynamics but we see that as soon as we turn the app intervention off, we increase lap times by about 20 seconds (and about 10%). In other words, the intervention is saving drivers time that can equate to an economic benefit. We note that this 10% has a confidence range (4% to 16%) and that we believe the true benefit will be higher when applied to non-professional drivers (see my comment on this below). The proportion of narrow road to normal width road, the number of narrow sections, and the density of cars are all important factors at play too.

The professional drivers at UTAC (and Damian and Marcus)

Another important dynamic is reducing head-on meeting occurrences, which is about safety rather than time. Being able to see round corners is as much about reducing stress and improving safety than simply saving time.

Head-on occurrences will occur if 2 vehicles traveling in opposite directions enter a narrow stretch at the same time. We will need to investigate this occurrence on different roads to estimate how frequently this occurs, but we observed significant reduction in this risk in our testing when the app was intervening. The hope for the MVC software is that these high risk situations can be fully prevented, similar to when traffic lights are placed on small narrow bridges, but with a far lower infrastructure cost.

Next steps

We are starting to plan for our next testing, which will be at Assured CAV (HORIBA MIRA) at the end of February. We will increase our testing to include 20 vehicles and include new road layouts.

We are also now exploring on-the-road use cases and are looking at events and festivals based in rural areas. Watch this space for more information!

A few modifications for the track setup at Horiba Mira are required. We will need to ensure the drivers can’t see ahead on the road before the narrow stretches, to truly replicate rural roads with ‘blind spots’. We will also need to split the efficacy assessment for time savings (often more than 2 cars) and head-on safety risk (2 cars).