For the past six years, I have spent what is most charitably described by my friends as an “embarrassing” amount of time and money making a crappy car slightly less crappy so that I may compete (and inevitably lose) a weekend long endurance race against other $500 cars in 24 Hours of Lemons. Everyone has their hobby and personally, I would only be embarrassed if my wife finds out what I truly spend every year on this $500 car.
When I started, I was constantly searching for ways to keep the car functional on the track for the 8+ hour days within the rules and budget. For example, back when the car still had an automatic transmission, I repurposed the A/C condenser into a giant “free” automatic transmission cooler.
Lately however, most of my “R&D” time has been allocated to figuring out how to adapt proper sporty aftermarket parts originally designed for other cars. The trick here is to properly camouflage them with just the right coat of paint/grease/dirt so they look “stock-ish”. Surprisingly, no one makes proper racecar suspension/braking components for 1995 Ford Thunderbirds.
Seven years after we started racing though, it doesn’t matter anymore if the car is no longer $500. We just want to race and the judges have already told us that we are a hopeless team. While I think of 24 Hours of Lemons as 75% racing and 25% theater, I suspect my wife tolerates this indulgence of mine because she thinks it’s more like 75% theater and 25% racing. While we most recently won “Organizer’s Choice” at the 2017 Sears Pointless Race, it was clearly for the thematic part of this series and not for our driving skills.
One thing that I can’t install a “cheaty” upgrade for in my racing car is the driver. As an endurance racing series, I cannot drive the car by myself the entire weekend. After an hour and a half on-track, it’s hard to avoid mistakes that will cost you repair time in the pits. I’m also too poor to hire a pit crew. So, suckering my friends into helping me with this hobby in exchange for seat time is a suitable alternative.
With far more time than any of my friends invested in what is pretty much a worthless car, I’m always looking for new solutions. One that really helped was to place a car horn right in front of the driver wired to an Arduino that gets triggered if the oil/coolant temps go out of spec.
Another experiment, which did not have as splendid results, was placing a live chicken in a coop on the rear trunk like R2D2 with a tiny helmet mounted camera on his head so he will squawk and let me know we are about to be hit by another car. For a fleeting second, the idea of replacing all of us amateur idiots with SW did have some appeal. I’m sure that a CEVA-powered ADAS system would be far less messy and far more PETA friendly than our first and only chicken teammate, who is now retired and living a car-free life in a friend’s backyard.
On the other end of the budget spectrum, a new autonomous racing series was showcased earlier this year at MWC by Roborace. These sleekly designed, driverless race cars are equipped with state-of-the-art hardware including five lidars, two radars, and six cameras. Sounds pricey, right? Well, I think that it’s a safe bet that it won’t fit in the $500-Lemons budget. These cars are designed to reach the upper limits of speed for thrilling entertainment, and no costs are spared. Theoretically, the bigger hope is that the innovations from the track will eventually make their way into road cars.
Maybe these algos might be relevant for certain scenarios in the developing world. The way taxi drivers leaving the Mumbai airport jockey for position, go into the corner five abreast on a road with only three actual lanes, and close each other out while aiming for the exit is something I’ve only seen elsewhere in a Lemons race. But what about technology that can make today’s street driven vehicles smarter, safer and more efficient?
My daily driver car, a Toyota Prius, is already more like my washing machine than my track toy. Obviously, neither the budget for anything I could cobble together in my garage for my hobby, nor the virtually unlimited budget of the self-driving Roborace cars would be a realistic template for a future self-driving version of my commuting appliance, which is what the car really does feel like.
A cost-efficient, low-power solution is the only way to scale self-driving technology for mass production. Here’s a look at two solutions that might be implemented on the road (as well as on the track).
Collision Avoidance Using Convolutional Neural Networks
Collision avoidance is a crucial feature for providing a safe path for autonomous vehicles. One of the key inputs to a collision avoidance algorithm is to determine “where the car should NOT” drive – aka free space detection. Together with our partner, AdasWorks, we demonstrated this at AutoSens 2016 on a cost-efficient platform targeted for high volume applications.
The solution combines AdasWorks' state of the art convolutional neural networks (CNNs) with the CEVA-XM4 intelligent vision DSP, to deliver a cost- and power-optimized system. The process of converting, integrating and optimizing AdasWorks' pre-trained networks to run on the CEVA-XM4, was made simple and fast thanks to the CEVA Deep Neural Network Software Framework (CDNN2).
Truck Platooning to Minimize Drag
For those of you not already familiar with racing, one popular way to conserve fuel is to drive together in a close formation to minimize drag. You see this in cycling races like the Tour de France as well as in car racing to save fuel and/or launch yourself ahead of competitors. Riding behind someone else for this purpose is known as drafting. Here’s an example of it in this video clip from the film Days of Thunder.
While this wouldn’t necessarily be valuable for street vehicles, it could save fuel for autonomous commercial trucks. Multiple vehicles going on the same long-haul route could use drag minimization algorithms to save fuel. Of course, if these trucks also happen to be talking to each other with CEVA powered M2M basebands (vs trying to knock each other out of the slipstream), I’m sure it would also would work much easier.
A Cost-Effective Solution for Commuting, Commercial Shipping and Ridesharing
Maybe there will always be demand for preposterously expensive systems when money is no object, as well as cases where a near-zero-budget will lead to somewhat ludicrous inventions. But those are just the extreme cases. Most self-driving cars will serve normal daily use cases like commuting, commercial shipping and ridesharing.
The best solution for these cases is the middle ground – accessible machine vision and M2M to enable mass-production. Meeting this demand will require efficient, programmable solutions that can be adapted quickly to evolving algorithms and neural networks to reach the market quickly and stay up to date with the changes and developments. Of course, if for some reason one of these Roborace chassis pops up on eBay for $500, I wouldn’t hesitate to push the ‘Buy it Now’ button first and figure out how to get the 24 hours of Lemons judges to accept my vehicle later.
About the author
Gunn Salelanonda is West Coast Regional Sales Manager for CEVA in the US where he’s been successful in licensing CEVA’s entire product line. When not licensing IP or chasing down his son Gil, he’s been far less successful as team captain & owner of Myopic Motorsports, a race car team campaigning in the 24 Hours of Lemons.