Invision AI is a Canadian startup building the first autonomous railway system combining visual AI and edge computing, looking to expand its solutions beyond the transportation space. The company was established by Karim Ali in 2017 and was invested by Fontinalis Partners, TA Ventures, Ontario Center of Excellence, Jayant Group, and TechStars. DIGITIMES had an interview with founder and CEO Karim Ali, who shared technical insights on edge AI and his vision for the company.
Q: Could you give a brief introduction and what inspired you to create this company?
I have a technical background with a Ph.D. degree in machine learning and computer vision from the Federal Institute of Technology Lausanne in Switzerland and worked at UC Berkeley and Harvard as a research fellow on computer vision and AI. Around 2013~2015, AI turned towards deep learning, which led to very huge, overnight increases in inaccuracy. We got systems that would be 50-60% accurate to systems that would be accurate in the high 90%s. It got to a point we realized that could actually lead to practical systems that solve real problems in the real world.
To deploy AI systems, we need to make sure that they're scalable. This means that it must be on device. We were looking for edge deployments of AI. And it also means that the type of information extracted must be actionable. And these are the two big problems that I saw.
On one hand, you have super high accuracy, which is great. But to get that accuracy, you need a lot of computing power, that's not really scalable. And the type of insights that are being extracted is not always very actionable. So, we set out to solve these problems, such as how to take AI and make it scalable and be able to build real-world products that solve real-world problems in a scalable way.
Q: You have adapted visual AI along with edge computing. What are the specific use cases that you would like your solutions to be implemented? Or what real-life problems you would like to solve for your users?
We see the potential for solving problems using AI and computer vision, like using cameras, radar, and LiDAR, and there's no end to the number of problems that can be solved.
We have three kinds of main use cases. We're looking at forward collision systems for trains, similar to the L2 or L3 system for auto but for trains. We have intersection monitoring systems. This could be a street intersection, or a highway intersection looking for cars coming in the wrong way, a pedestrian on the highway, stopped vehicles, etc. We're looking at level crossings, such as where trains meet cars. For safety, our system can say, hey, there's a vehicle obstructing the path of the train; keep the barriers up.
And then we're also looking at vehicle occupancy detection, such as counting people from the roadside to automate enforcement of high occupancy vehicle lanes.
The key for us is, that we are building a platform that is reusable across a lot of these applications. And the way to make that platform reusable and scalable is by hitting three important points:
One is edge, everything must happen at the edge and not on the cloud. That way, you can keep the data close to the field and close to the sensor, and don't have to ship it to the cloud, so there are no privacy and computation concerns.
The next big point to enable the flexibility of a platform is collaborative sensing. You need to have multiple sensors or multiple cameras collaborate together to produce a cohesive understanding of the world.
And the third point is really extracting actionable measurements from the system such as geo-locations, sizes, speeds, headings and so on to produce a digital twin. In the end, Invision AI is about an on-device digital twin.
In our computer vision AI, when detecting vehicles and other people, we don't just detect these people, we geo-locate them. So we tell you where they are in the real world. What are the GPS coordinates of this vehicle? What is the length and width and height of this vehicle? How fast is it moving? Which way? Is it oriented? Where is it going to be in the next few seconds? And, you know, the combination of on-device and these collaborative cameras, collaborative sensors, and extraction of digital twin information, that, for us is the key platform to be solving many, many, many use cases.
We're very focused right now on intelligent transportation because this is where our solution shines the most, where there's the greatest need for optimized digital twins, essentially. But we believe that the applications go well beyond just the transportation space.
Q: You have done several autonomous rails projects successfully with two companies. I'm just wondering, why pick railway to deliver your first smart mobility solutions?
Part of the reason is technological. We believe the first autonomous vehicles are actually going to be in the railway sector. There's a lot of talk and a lot of hype around autonomous cars. But they have many problems to solve because it is a much less controlled environment than the railway environment.
In rail, because there's a single operator that controls trains and level crossings and train stations, you get to look at the problem from a cohesive standpoint. Most operators who are interested in the railway sector have already figured out that you need to have autonomous train systems working in conjunction together and collaborating with fixed infrastructure systems that would be installed on level crossings.
We've started to build those systems working together from day one. That's very exciting. The autonomous car industry, for four years, they've been saying all autonomous cars will work on their own. People are realizing, "Wait a minute, most accidents are happening at intersections." If we had a cohesive coverage of the intersection, that's a lot of information that can be used.
An autonomous car with all of its sensors can only see so much. If you had an eye in the sky that could see a danger well ahead of time could send a warning well before pain, such as "You should break now!" when there's a problem happening at that intersection in the next five seconds.
We were lucky to have had a very wonderful collaboration for the past two years with Thales, a leading technology integrator. We're working with Metrolinx, which is, the second-largest operator in North America, they have a deep commitment towards safety, which is why they're there working with us on forward collision warning.
Q: It's very inspiring to listen to your sharing especially when you mentioned that with coordinated IoT sensors on the road, then we can prevent a lot of accidents. Well, if there is smart city planning covering all those systems, is that the vision for Invision AI to implement in the future?
We want to be a platform that can be leveraged for all kinds of applications. Yes, that's 100%, where we would like to go. When you look into the future, we strongly believe that, and this is what we're building, moving sensor suites working together with fixed sensor suites, fundamentally, it's going to be the same technology on both the same type of AI algorithm detection and tracking, that are running on both and working together. To enhance safety, throughput, reduce congestion, just build a better and safer transportation system.
Q: In what way, can Taiwanese companies collaborate with you? Are you looking for partners or hardware partners that you can work with?
We are looking for partners for a safety-driven application, like low latency communication between these cameras, or between the computing nodes of each camera. The kind of hardware for a collaborative multi-sensor system can be deployed without any wiring -- wireless, low latency wireless, 5g capabilities, so we can provide a turnkey solution. Taiwan has integrators, and hardware producers, so, always looking to see what kind of relationships we can form.
Q: What is your fundraising status and plans for fundraising this year?
Our last fundraiser was in mid-2019. And we've been heavily bootstrapped since then. Currently, we're financing the company through revenues and have managed to grow the team to 20 people. With this model, right now we feel the time was right, for another fundraiser. So we just started a fundraiser, which we're hoping to close in Q3. And now we're primarily looking for strategic alignment. We're targeting US$5-10 million.
Q: What kind of companies you would like to have this kind of strategic alignment, for example, are they the ones that complement you?
Very possible. Could be camera companies, edge computing, in the intelligent transportation space, and the like. We'd be happy to have these discussions. We have several partners already, but we're always looking to grow with partners because we see our market as a global market.
Q: And do you have any plans to expand your markets overseas this year?
Yeah, definitely. Wherever we can get traction is where we're going. We have projects in countries that were already in Greece, France and Switzerland, Canada, the US, and Israel. The Asia Pacific region is also a very key region for us because many countries there are very advanced in the smart city and intelligent infrastructure, and transportation space. For APAC, we are working on expanding our markets to South Korea, Japan, Taiwan, Singapore, and Hong Kong.
Series A Targeting US$10 million in 2022
CEO and founder Karim Ali
CFO Craig Loverock
CTO Carlos Becker, VP Product Emanuel Corthay
Toronto (Canada) and Lausanne (Switzerland)
Invision AI founder and CEO Karim Ali
Photo: Invision AI