Road trips with an electric vehicle (EV) are often described like driving with a toddler - the car decides when and where it wants to eat and rest, and the driver obeys.
However, Tesla still happens to be unrivaled when it comes to the overall sense of security on long trips. Its tightly integrated in-car GPS navigation seamlessly suggests which route to take, when, and where to recharge, making the hunt for chargers painless and freeing the driver from any route planning work.
In the series of previous blogs about EV navigation topics (Part 1, Part 2, Part 3) we've explained how AI can help build a highly accurate EV driving range prediction system. Tesla uses a proprietary system for that purpose and its implementation details are not transparent. At the same time, most other EV car manufacturers use physics models for range estimation, which may fail miserably, leading exactly to why the term "EV range anxiety" was coined.
Our aim was to create an accurate AI-based EV range prediction model which could be used on virtually any EV vehicle with the same or higher range estimation accuracy compared to the benchmark Tesla technology.
So, did we succeed, and how our AI-based model compares to Tesla's?
Here's the story from a real-life trip we've made.
This winter, we made a 1,000+ km round trip from Zagreb, Croatia, to Maranello, Italy, with our Tesla Model 3. Specifically, we wanted to benchmark our EV GPS navigation to this leading ecosystem.
The very same route to Maranello we’ve driven numerous times with a gas car. Now, we wanted to see whether we could make the journey as similar as possible to a gasoline car drive - to reach the most negligible overhead of charging time and, naturally, not run out of battery.
Tesla Model 3 obviously cannot make this 570 km one-way trip without stopping to charge. We have to map out charging pit stops in advance, and we can summarize our requirements on route and charger selection as follows:
We want to spend as little time as possible on charging, aiming at 20-30 minutes per charger; therefore, we prioritize fast chargers.
We want to charge exclusively on the Tesla Supercharger network. Unlike standardized gas-fuelling nozzles, EVs have four main charging plug types, and carmakers typically support only one or two. However, if not possible, we're more than willing to charge on any other charger with a compatible plug if push comes to shove, even if it's a slow charger, as long as we keep the engine running.
Our Tesla has a 423 km advertised EPA range (range measured on 55% highway and 45% city driving), which based on our driving experience, translates to approximately 250 km of highway driving in cold weather. Achieving the advertised range requires perfect conditions, including temperature, driving speed, road slope, etc., and realistically one can never get it all perfect.
Therefore our 570 km one-way trip would need at least 3 charging stops, possibly even more. Although it might seem feasible to make the entire trip with just one charge, it's only sometimes necessary to charge the battery fully. Charging from 20-80% takes roughly the same time as charging from 80-100%, so we'll aim to keep the battery in that range. This practice not only extends its lifespan, but also speeds up charging time
Of course, these are the ideal requirements. We’re aware that the situation on the road can dramatically change, and we’re prepared to make compromises. Being left on the highway waiting for roadside assistance is not an option.
To make charging at our destination as easy as possible, we'll research charging options and pre-download any necessary apps.
The mapping out of the optimal charging stops along a route is time-consuming and laborious if done manually, which is why most drivers rely on in-car GPS EV navigation or third-party EV route planner applications. Moreover, the use of GPS navigation offers flexibility in terms of updating the charging plan and rerouting, making it a preferred option for many drivers.
To make all of this possible, a route planner or a GPS navigation must have a comprehensive understanding of how the car’s battery discharges on each trip segment. The precise estimation of battery consumption is the cornerstone for determining the locations where to charge.
For example, the battery depletes dramatically faster in cold weather, on uphill drives, while speeding, having the AC or heating on, etc., and the route planner needs to be aware of these variables. Unlike 3rd party route planners, in-car EV GPS navigation has a distinct advantage because it has direct access to the sensor data, which can be incorporated into the battery consumption calculation.
Quality in-car EV navigation suggests a route that minimizes charging overhead, continually reevaluates the charging locations, and timely warns and instructs extra charging if necessary.
This 1,000 km trip we've driven following Tesla’s navigation suggestions and simultaneously compared it with Mireo‘s proprietary GPS navigation for EVs. Mireo‘s EV navigation leverages car sensor data to provide accurate estimates of battery discharge for each trip segment, allowing for optimized charging recommendations.
Similarly to Tesla’s, our EV navigation employs machine learning algorithms (ML) to learn how an actual car driven by an actual driver consumes energy in different driving scenarios and adapts to changes in driving style and vehicle and battery characteristics.
The results of our 1,000 km round-way trip to Maranello were stunning:
Excluding the battery preconditioning, our prediction of the remaining battery capacity at the end of each trip segment did not exceed the actual battery capacity by more than 1%.
More precisely, the first estimate of the remaining battery, calculated at the trip start, didn’t exceed the actual value by more than 1%. In contrast, Tesla's estimate of the remaining battery heavily oscillates up to 5% of the battery capacity during the first ~30 km of driving and then settles at the final value. Our estimate proved to be accurate from the start.
The only segment in which our EV navigation underperforms is when battery preconditioning is activated. Namely, when navigating towards a charger, Tesla’s navigation instructs the car to heat the battery to the optimal temperature to speed up the charging time and protect the battery, consuming up to 10-15% of the battery. Our EV navigation, effectively as 3rd party navigation, is entirely unaware of this process, making it impossible to account for realistically. As a result, all of our tests were conducted without battery preconditioning.
Let’s finish this blog with a few trivia from our trip to Maranello. The original plan was to leave Zagreb with 100% battery, charge to 100% in Ljubljana, recoup in Palmanova, and charge one last time near Ferrara. However...
As Teslas are cloud-based (connected) cars, the lack of internet connection can cause various challenges, to say at least.
Misfortune #1 - Lost cellular data connection
As our adventure unfolded, we encountered the first mishap - a lost cellular data connection while charging at the Palmanova supercharger. Because of this, our Tesla app stopped receiving battery charging updates! We had to rely on our experience and gauge that 20-ish minutes of charging would suffice. Except it almost didn’t because of our next mistake.
Misfortune #2 - the wrong highway exit
Driving south towards Occhiobello Supercharger with only a little battery left, we accidentally took the wrong highway exit, which nearly proved disastrous for us. While a common mistake in a gas car, it can mean a complete change of plans when driving an electric one. Suddenly we were left with not nearly enough remaining battery to safely make the detour and arrive at the charger we were headed for. We had no choice but to head in the wrong direction, towards Vicenza on the west, to reach the nearest charger.
Thankfully, the trusty GPS navigation saved the day by rerouting us to the nearest charger. However, in cases like these, driving without GPS navigation in unfamiliar areas is a disaster waiting to happen.
Mireo's EV battery discharge model calculates the furthest point a vehicle can travel to, along any possible route from the current position
Several factors contribute to the popularity of Tesla cars. The key of them is Tesla's success in creating an EV driving experience that rivals that of traditional gasoline-powered cars, achieving this feat much earlier than other carmakers.
Tesla has also recognized the importance of developing a proprietary network of chargers and a reliable route planner, which has greatly reduced the range anxiety associated with long-distance travel in electric vehicles. Tesla owners have far greater confidence in the presented battery prediction and available range information.
Through its tightly controlled and closed environment, Tesla has built a loyal fanbase of satisfied customers who appreciate the seamless integration of the EV navigation system with the rest of the car.
Our objective was to provide Tesla-like EV GPS navigation with a highly accurate range estimation method to virtually any EV vehicle manufacturer. To achieve that goal, we've developed an AI-based system for estimating the remaining driving EV range, which "learns" how an actual vehicle driven by an actual driver consumes electrical energy in various conditions and on various routes. Compared to Tesla's proprietary one, our method yields not only comparable but in most cases, even better predictions.
Paired with the appropriate upgraded GPS navigation that includes EV-specific features, the system provides a Tesla-like experience on long journeys that require at least one extra charging stop.
It is worth noting that we've used literally just a dozen CAN signals from our Tesla Model 3 to train range estimation AI model. In other words, to achieve accuracy that effectively eliminates EV range anxiety, we have proven that it is not necessary to build complex models that measure every single energy consumer in the vehicle. With only a few sensor values from the CAN, we can provide far better range prediction and driving experience to any EV vehicle on the market.
The in-car EV range estimation is every now and then off by a stunning 20-30% across the board.
Physical-based EV range predictions can't consistently provide accurate results.