There’s a race in Silicon Valley to build self-driving cars. The mission is safer roadways. All the startups are gunning to hit Level 5 for full automation. But we’re not quite there yet.
Most of the tech teams are currently straining to hit Level 4 autonomy (mostly autonomous).
The self-driving community in the valley looks like a mess of genius. There are numerous intertwined partnerships between startups and investors. Tech companies, mapmakers and automakers all intermix into the confusion.
Amid the chaos of interlinking collaborations, one brilliant team of engineers stands out with a promising edge.
In the self-driving race, Drive.AI has been generating a great deal of attention. Their team is uniquely composed of multiple deep-learning experts straight from Stanford University’s Artificial Intelligence Laboratory. Their scalable approach and aggressive pace makes their project designs quite promising
Their claim to fame?
Drive is the only self-driving startup to birth designs with a thoroughly holistic use of deep learning technology. By focusing on humanized vehicle learning from the beginning, the future “magical world” of self-driving cars can be safe, friendly and reliable.
When it comes to creating effective self-driving cars, the biggest challenge has always been predicting the unpredictable. With a never-ending variety of potential road blocks, this has been especially difficult for technicians to account for. Researchers are working to gain excessive roadway data to better predict these obstacles.
With infinite potential road blocks, it’s crucial to develop safe predicting systems. Most self-driving groups are continuously increasing obstacle-avoiding algorithms based on newly understood obstacles and rules. They then program these new rules into the vehicle. This is a complicated process. It makes environmental learning a constant struggle.
Deep learning is better, and it’s had tremendous success in many nuanced situations. For instance, familiar machines such as Siri and Alexa have been able to master voice recognition. That’s a common example of deep learning responsiveness at work.
What’s unique about Drive is the way they have incorporated human learning abilities into the vehicle. They birthed their models with fully integrated deep learning use. This means that the machine learns its environment similar to the way humans process every day information.
The system starts to learn for itself and begins to create its own rules.
Humans learn more from familiarity recognition, while robots process new information using programmed algorithms. Think about it. If you show a child an image of a cat, that child will easily be able to identify it as such. We know it’s a cat because we’ve seen tons of cats before. We don’t have to break it down. We just know. That’s how human brains work. A robot would break down the cat photo piece by piece in order to identify it: two eyes, two ears, mammal, colored fur.
From day one Drive has incorporated human learning mechanisms into their design. They have been able to create what we call “responsive algorithms.”
This means the vehicle won’t have to be trained for every imaginable situation. It has the ability to recognized an old lady in a wheel chair and a child playing with a ball under the same category of pedestrians. This is possible because of the camera image understands the perceptual pattern. With that information, there are decision making and motion planning patterns as well.
Imagine a four-way stop where other drivers are supposed to follow specific rules. But sometimes they don’t. With situation-based deep learning, the vehicle can make a safe decision. This element of deep learning contrasts with the more traditional, rule-based systems.
Drive has designed their tech processes to make decisions more closely related to the way humans process new information. With this type of design, they’re able to scale it into all kinds of challenging environmental situations.
Deep learning AI works great for image recognition (a technology we’re familiar with from Facebook). As it begins to classify information, the AI recognizes patterns. Even further, it can then apply those patterns to classify objects it has never been exposed to. Drive believes deep learning is the most effective form of AI, and that it can be applied to the pressing issue of self-driving cars.
Photo courtesy of Drive.AI Launch Squad
From your iPhone to your refrigerator, machines are getting smarter. With all this accepted tech progress, autonomous driving is still an incredibly new form of AI on the market.
The group of former lab mates believe that deep learning in autonomous vehicles is the future of transportation. Most self-driving groups will use deep learning to solve specific problems. However, drive started with deep learning as their basis of operation. Their complete integration goes against the traditional robotics approach.
Less than a year after they entered the public scene, Drive has sent a fleet of four nearly autonomous vehicles navigating through the streets of San Francisco Bay Area.
Their mission statement claims they’re “building the brain of self-driving cars.”
After attaining their Autonomous Vehicle Testing Permit, they’ve prioritized safety throughout all of their trial runs. By mid-February they released their first public video of their self-driving technology. You can clearly see their system navigating the roads in Mountain View, Calif.
The video displayed their test vehicles operating through rain, darkness, and even hail. The robots they’re building have been able to smoothly navigate these especially challenging weather conditions.
Drive works to develop systems that quickly interpret data from sensors and learn to control the vehicle’s behavior. More specifically they’re trying to classify the norms of driver and pedestrian behaviors. They anticipate that human behavior will change around self-driving cars.
In order to train these types of systems, extensive data is needed. For most startups, the goal is to gather as much information as possible for a vast array of situational variety. Developers then have to take all of this data, interpret it and do something useful with it.
Drive.AI also specializes in gathering high-quality data. They process and annotate the data specifically for deep-learning usage. It’s a laborious task to note all the options captured by self-driving car sensors. To put that in perspective, most self-driving startups are spending 800 human hours annotating for every one hour of trial driving time.
The Drive.AI startup has been able to streamline this tedious task using deep learning-enhanced automatons. Their relatively small group of annotators spends the majority of their time training systems for new scenarios. It’s more of a validation step at that point.
Every year 33,000 Americans die in automobile accidents. Around the world that number ramps up to 1.2 million. The most unpredictable and dangerous element of roadways has been human behavior. Human error causes 90% of all traffic accidents.
We haven’t learned to interact well with these machines yet, which is to be expected. To accommodate for this learning curve, Drive has been designing ways for the vehicles to clearly communicate with other drivers and pedestrians.
The focus is on incorporating human interaction as a key strategy in vehicle communication. They are extremely sensitive to the impact of these vehicles on the people around them.
When it comes to communication, humanity has created an endless stream of understood nuances. Consider our developed system for conducting necessary information to other drivers and pedestrians: nods, hand waves, eye contact, friendly honks… and less friendly honks. It’s an interaction rhythm that’s nearly impossible for robots to emulate.
Now Drive.AI is working on a whole new language that focuses on trust and transparency. They’re developing effective ways to communicate through displays on the vehicle. They’ve experimented with using text, sounds, lights and even some motions to interact with drivers and pedestrians.
The goal is to re imagine the relationship between cars, people and the world. Drive wants to develop ways to help people understand what the car is planning to do beforehand.
The end result of Drive’s mission is to give people hours and hours back of our time. We can enjoy living instead of driving.
The average American commuter will spend 42 hours stuck in traffic each year. All of this congestion leads to poor physical and mental states. Our vehicles add so many issues to our daily lives. That’s all outside of accidents. From air pollution and wasted time to sucking your income and making you fat, our current transportation is far less than ideal.
Self-driving vehicles can empower the disabled and the elderly. They can transform landscapes and cut back on CO2 emissions. They can even make cars fun again.
People have been dreaming about improved transportation for many decades. The self-driving race gets faster and faster, and each company wants to be first. News stories are full of confident press statements about grand visions and nearing completions.
When will our streets be filled with a majority of autonomous vehicles? With the brilliant engineers on board, we’ll continue to make progress. The predictions range from 2020 to 2040 for full integration of autonomous cars. We don’t know. But we hope that we will be prepared enough to see this beautiful world change.