From DARPA to TONY – The (Autonomous) Journey of Perrone Robotics’ CEO Paul Perrone
TONY is short for ‘TO Navigate You’. It’s a vehicle-independent, retrofit kit for use in the autonomous transit of people and goods. And it’s been developed by Perrone Robotics, a U.S. provider of fully autonomous vehicle systems. Its founder and CEO is Paul Perrone, who’s been in the AV technology space for over 17 years.
Auto Futures has been talking to Perrone about his company, its innovations, and a journey that’s taken him from MAX to DARPA to TONY.
The 2005 DARPA Grand Challenge was a race across the Mojave desert with a $2 million prize offered for anyone who could build a driverless vehicle that could complete the 100+ mile desert journey. Perrone’s team entered the race with a vehicle that featured ‘MAX’, an AI-powered autonomous platform.
“The stakes were high, but the opportunity to prove MAX’s mettle was irresistible. After forming a small team, a vehicle named Tommy, embodied with MAX, advanced through numerous application trials and was invited to the 2005 DARPA Grand Challenge events in the west,” regales Perrone.
“After a showing there and in the subsequent 2007 DARPA Urban Challenge, it was clear that MAX was proving itself as a viable full-stack autonomous software platform for complex autonomous vehicle (AV) applications.”
After a few years in R&D mode developing the foundations for MAX, in 2008 the company set out to begin commercial deployments of MAX operating in a wide variety of vehicle types and in industrial applications.
The company even collaborated with rock star Neil Young, automating his 1959 Lincoln Continental conversion to an electric vehicle, called LincVolt.
After a 2016 investment led by Intel Capital, the company expanded and began a series of strategic projects in the automotive and industrial space working with Fortune 500 companies, automotive OEMs, and tier 1 suppliers.
By 2018, with MAX at its core, Perrone Robotics had fully developed a ‘drop-in any vehicle’ retrofit kit approach to autonomy that it began bringing to market in the deployment of fully autonomous shuttles and vehicles.
“With over 300 person-years invested, over 37,000 AV miles travelled, and over 30 different vehicle types outfitted, we are pleased to be bringing our proven turn-key TONY shuttles and TONY retrofit kit, with new vehicles, to market and into production for the proliferation of autonomous transit NOW,” says Perrone.
State-of-the-art Autonomous Testing Facility
The company has a patented approach for verifying and validating system safety in autonomous vehicles (AVs). This novel approach involves an onboard ‘safety watchdog’ subsystem which monitors the complex autonomous drive system for hazardous events, says Perrone.
The company can now boast over 37,000 automated miles of travel, 30 different vehicle types outfitted, and testing since its inception.
Current pilots include a Fortune 100 logistics company application of automated tractor trailers, all electric neighbourhood electric vehicles at a U.S. Army base, and an all-electric transit van for the Jacksonville Transportation Authority.
“The Jacksonville Transportation Authority (JTA) led by Nat Ford have a bold vision for the future of autonomous transit. When we won the contract, we were excited to provide the world’s first fully autonomous, all electric, ADA compliant, FMVSS compliant, Altoona tested, with Buy American compliance on path transit van – the AV Star. The AV Star is able to travel on public roads with a maximum speed of 65 mph,” says Perrone.
“This project has significant implications for all transit authorities as the industry reduces their carbon footprint and looks for safe, economically viable solutions. Our ability to retrofit any vehicle – is an important feature to fast track those initiatives for transit authorities across the nation,” he adds.
Through its partnership with Capstone Holdings Inc., the company has announced the opening of a state-of-the-art autonomous testing facility in the American Center for Mobility (ACM) in Michigan. Perrone will have scalable garage space, a test track and dedicated time allocated for the continuous and varied scenario testing of its AV technology on various vehicles in various conditions.
“The ACM test track will bring an added level of shakedown testing to its QA and test repertoire. Perrone will collaborate closely with faculty and students from Eastern Michigan University on site at ACM in the transition from R&D to practice of advanced next generation AV technologies. This is a great partnership for tech-transfer and provides opportunity to seed our future workforce,” explains Perrone.
Verifiable and Validate-able Approaches
Perrone believes that it will be 15-20 years until fully autonomous vehicles will be on U.S. public roads and at scale. He says many projections from large players have been off base, as they were not able to meet the timelines and/or overcome some of the technical challenges.
“The reality is that many of the current approaches, which rely heavily on deep machine learning techniques, that are being undertaken by some of the big players and small players, were/are fundamentally flawed. I believe that there are many initiatives underway that are more realistically categorized as titanic R&D efforts which dabble in probabilistic and unpredictable AV development more so than hard track and deterministic AV solutions, like those Perrone offers, which will be fieldable,” says Perrone.
“As a result, I think the predictions for on road commercial autonomy are way off base right now. And many deployments that overly rely on neural networks and more bio-inspired approaches to autonomy, if deployed, will be catastrophic and potentially set the deployment pace back decades.”
“In the meantime, we’ll be focused on the deployment of immediately useful and fully autonomous vehicles using more deterministic and verifiable and validate-able approaches in geo-fenced applications this year,” adds Perrone.
“ When there are safe, solutions out there and realistic timelines, then we all win.”
Perrone has very real concerns about the way driverless transport is travelling. He wants to see autonomy leveraged in a practical and safe fashion because the technology offers very real benefits.
“This can save lives now, can improve lives now, and can produce real and tangible benefits for companies now. We don’t want to spin wheels and waste time on trying to do the equivalent of trying to colonize Mars before we’ve made and demonstrated over time a safe and reliable rocket that can get us to the Moon and back,” he says.
“The market and many technology innovators, may feel its easier to think you can train a system to learn how to drive a vehicle autonomously. (Intellectually this is the new shiny gold object that attracts many an engineer into its grip.) On top of that, many large organizations are touting it as competitive advantage. But I feel, and in fact I know, that it is a trap.”
“Why? Because these predominant deep learning approaches leverage neural networks, which are good at pattern recognition, but not so much in a broader application of making decisions on behavior of how to drive the vehicle under the nearly infinite number of scenarios that can be countered.”
“As a result, my concern is that the pressure to deliver this type of autonomy may supersede safety and proper testing – especially when it comes to scale. I worry that many a vehicle will be launched and that a looming catastrophe is on the horizon as they’re put to the test at even a moderate level of scale.”
Perrone likens the divergence in approaches to autonomy to the early days of flight.
“Many bio-inspired approaches to flight resulted in flapping winged contraptions that came crashing to the ground. However, the Wright brothers took a ‘controls first’ approach and focused on the more deterministic stability of flight. Similarly, the deep neural network based approaches to autonomous vehicles right now are themselves bio-inspired approaches to creating an artificially intelligent driver. Whereas the approach we subscribe to is a ‘controls first’ approach that is more deterministic.
“The unpredictable and unknowable decision making logic that is being baked into vehicles that will be transporting people to work and children to schools, has me deeply concerned. This simply can’t happen and it’s the path being taken by many companies with billions of dollars of backing, so they’re on a path that can’t easily be reversed because of the vested interest in the direction.”
“I fear that under this path, the landscape for autonomy in 2030 will look just like it looks like today. However, with more deterministic and controls first approaches, and a strategic approach to roll out to simpler domains first, we can begin to reap the benefits of autonomy now without being read its horrors,” concludes Perrone.