4 ms·

Hi Tom, I've done a lot of work with autonomous self driving robots. Specifically relating to your company is the research I've done on an autonomous wheelchai

Hi Tom, I've done a lot of work with autonomous self driving robots. Specifically relating to your company is the research I've done on an autonomous wheelchair that can navigate a sidewalk between two locations. I know it's blatant self promotion, but I would encourage you to take a look at some of my lab's papers [1], which specifically focus on building the kind of system you describe. One question I have for you is about your business model. Do you plan on selling these to a restaurant directly? Or do you plan on building a fleet of these things and licensing them out to restaurants (or basically any delivery service)? The answer to this question really informs your value proposition and the technology you need to develop. For instance, you have emphasized that you've done some teleoperation tests, but this requires a human in-the-loop, whereas the value proposition for this devices is in removing the need for a human. If I have a human operating this thing, I might as well have him go out and deliver the food. So this points to the need for an autonomous system. To really build a scalable business I think the best route would be to focus heavily on building the robot system. Do you have someone on your founding team who is a roboticist? You say you're looking for engineers, but I would say you need someone on the founding team who knows their stuff. Here are some technical challenges that are still active research areas that you'll need to solve to make this an attractive option: - Navigating without GPS. GPS is not accurate enough in cities or specifically on the sidewalk - Large scale mapping. How do you make and store maps? You also have an opportunity here to use the delivery robots to constantly update the map and inform other robots of changes. - Long-term autonomous operation. How do you recover from lost localization? - Autonomous operation in different environments. e.g. indoors vs. outdoors - Sensor calibration and characterization. You have an asus 3D camera on your demo car, but these fail in various lighting conditions and have limited range. You'll need better sensors. - Crowd navigation. How do you predict the behavior of humans? How do you route a path through a crowd, a seemingly impenetrable obstacle, but one humans can navigate easily. - Crowd localization. How do you figure out where you are in a crowd, when you have no reference to anything else? And much much more. Anyway, best of luck! It's a problem I've put a lot of thought into, so feel free to drop me an email if you have any questions. [1] http://vader.cse.lehigh.edu/publications.php http://vader.cse.lehigh.edu/publications.php

tomjacobs 11y ago
Thanks Corey, Yes indeed GPS sucks on the sidewalk. Outdoor feature-based SLAM is the way to go. Congrats on getting the robot to go navigate 12K by itself with no loss of localisation. The human operator is the backup, similar to Starship. A separate mapping operation needs to be done first to pre-learn routes. Recovering from lost localisation is tough and will need people to go recover the unit. The Zed 3D camera operates better outdoors than the Asus, and solid state lidar sensors like Quanergy's are coming to market at a rapid pace and lowering costs. http://spectrum.ieee.org/cars-that-think/transportation/sensors/quanergy-solid-state-lidar http://spectrum.ieee.org/cars-that-think/transportation/sens... Crowds are tough too. The robot would not want to move too much in a crowd, and perhaps is best for it to just stay still until the crowd passes and it can re-localise. These solutions and technologies are only going to get better with time, not worse, so it'll be a much easier problem given some short time.

tomjacobs 11y ago

Thanks Corey, Yes indeed GPS sucks on the sidewalk. Outdoor feature-based SLAM is the way to go. Congrats on getting the robot to go navigate 12K by itself with no loss of localisation. The human operator is the backup, similar to Starship. A separate mapping operation needs to be done first to pre-learn routes. Recovering from lost localisation is tough and will need people to go recover the unit. The Zed 3D camera operates better outdoors than the Asus, and solid state lidar sensors like Quanergy's are coming to market at a rapid pace and lowering costs. http://spectrum.ieee.org/cars-that-think/transportation/sensors/quanergy-solid-state-lidar http://spectrum.ieee.org/cars-that-think/transportation/sens... Crowds are tough too. The robot would not want to move too much in a crowd, and perhaps is best for it to just stay still until the crowd passes and it can re-localise. These solutions and technologies are only going to get better with time, not worse, so it'll be a much easier problem given some short time.