Ai4 recently hosted an “Ask Me Anything” session with Harish Skumar on our Ai4 Slack Channel. Read the full transcription below…

MODERATOR: Hello everyone! It’s a pleasure to welcome our next AMA guest Harish Skumar, Roboticist & Program Lead of Racing Vehicles, General Motors. Full bio here: https://ai4.io/digital-events/#filter=.category-102.  You now have one hour to ask him anything. Ready, set… GO! 

HARISH: Hello everyone! Looking forward to great questions and discussions.

PARTICIPANT 1: Hi Harish, thanks for being here! Has your team priorities changed at all in response to COVID-19?

HARISH: We are taking more precaution to how we interact and the frequency. We are also having fewer needed people while working on continuously working to become better.

PARTICIPANT 2: Harish, thanks again for being here! I’d like to throw a question into the ring as well. What was the greatest technical challenge when building the nuclear power plant inspection robot?

HARISH: The greatest challenge has to be the environment itself. The high temperatures and gamma radiation.

PARTICIPANT 3: Hey Harish, what methods do you use to bridge back the real world data to the model? Do you have a federated method to rejoin abnormalities from real life into your base model?

HARISH: We use virtual models to maximize the overall data that can be simulated as a base. Then, once we have the data from pervious designs, we take the model to real world testing to account for unknowns or variables of the given environment. Once we have the consistent variables identified, we update the model to learn.

PARTICIPANT 4: What is your approach to building or adding to a data/robotics team?

HARISH: My main approach would be identifying the needs of the data and/or robotics. When you have the needs identified, then you have a starting point. In my experience, we have started at a broad scope prior to have a fully defined and developed design. This approach allows to analyze inter-linked components and data to reduce complexity. The key is not to have too much data handled at once, optimizing the data and the robotics to allow easier troubleshooting/root cause.

PARTICIPANT 2: I’m not sure if this question is totally within scope, but what are some of more unexpected places you’re seeing AI being applied in the automotive industry?

HARISH: That is a great question! Besides the current race between electrification and EV/AEV even, AI is being applied to mapping. There is a new trend of using the very useful technology of GPS and AI interface to develop over-laying data. The GPS data is used to teach AI the constraints of the environment while allowing to learn the variables such as truck swerving out of it's lane due to sudden cross-winds. In the truck example, the AI know the route using GPS, and based on the external data such as from camera and sensors, is able to not only recognize the object as a truck. AI + GPS now can understand the environment.

PARTICIPANT 5: Hey Harish! Which technical aspects of your work on racing vehicles translate in everyday cars for consumers?

HARISH: There are a few technical aspects that are usually translated to consumers. These are the high performance trims with trickle down from the Racing side. Most common is the power, safety and efficiency. We are starting to translate getting more power with smaller engine volumes and driving to improve/maximize fuel range. In racing, it is very important to maximize the fuel volume standards that are set forth by regulation. This strict regulations drives racing teams to design and optimize all the power from an ICE (internal combustion engine) to the road as efficiently as possible. In a race this could mean needing to pit more often or even longer such as a 24hr race. LeMans type races are where these are key advantages. This does translate down to most consumer vehicle, more on the performance than SUV's.

PARTICIPANT 2: Again, not sure about scope here, but how soon will we see our roads and highways transitioning to >50% autonomous drivers? Are the largest constraints with the technology or other factors like infrastructure or legislation?

HARISH: I believe the two questions are interlinked. In order to have AV on road, we need to first address infrastructure legislation as one. First we need to have proper road markings on all roads, as AV are all based on visual and data. There needs to be a drive to standardize what is needed for AV to start increasing performance. Once we have infrastructure legislation which can maintain roadways and highways, then the larger roadblock will be safety, responsibility/liability in an event of accident. Insurance will be a new concept, who will be responsible? how do you go about identifying who's at fault in an incident, would it be the computer/AI or OEM? And lastly, consumer confidence. Tesla has started the lead and they are starting to make us think about what is possible, and also are we ready.

PARTICIPANT 2: Quick follow up. You don’t believe that technologically we’re getting close to a point where proper road marking won’t be a factor because the CV technology is actually better at understanding the environment than a human? Ie. If the roads are good enough for humans, then they’re exceedingly well equipped for machine?

HARISH: I do believe we are on the way, as one of the other questions earlier, AI application to mapping. As an example, when you are driving and you see a pothole and avoid it by driving around it or safe maneuvering. AI currently are not focused to see the potholes, so could result in a flat tire or bent rim. This not only is an inconvenience but a possible danger to the occupants. We humans still have a thing or two to teach AI, but we need to first make the AI perfect eve and avoid as we do now. That was the thought behind we are getting there, but still have long ways to go.

PARTICIPANT 2: Sorry to bombard you, but I always like to ask the vendor question. Are any AI vendors or solution providers that play a key role in your work with AI + robotics?

HARISH: In my experience, there are a few major players which I have encountered. Most of the vendors are in the AV space and less in the industrial robotic space. AI + Robotics is expanding and there are few applications such as KuKa, who have robots which can make basic decisions dependent on sensor systems. A tire manufacturing who makes multiple size of tire is an example, you might have one conveyor with various tire sizes, to a sorting robot of sorts. The robot now will have to identify the size of the tire, which conveyor that tire needs to go for further processing/packaging and correctly execute. AI + Robotics definitely has a play here, as they robot can be taught the basics of incoming to tires, but AI can enable it to learn as the company produces more various sizes without re-teaching the robotic side.

MODERATOR: If there are no further questions, then we’ll call it a wrap! Harish, thank you again for taking time out of your busy day to take the time to be with us and answer questions for the Ai4 community. It’s been great having you.

HARISH: Thank you very much for having me, this was a blast! Would love to get more questions and expand, this is a great way. It's been a pleasure!