News

Does Seegrid Hold the Key to Cheap Autonomous Vehicles?

April 09, 2016 by Aaron LaBarbera

There are plenty of companies entering the autonomous vehicle ring like Google, Tesla, and Baidu to name a few. But none of them currently have the operating know-how of a small Pittsburgh-based company called Seegrid.

There are plenty of companies entering the autonomous vehicle ring like Google, Tesla, and Baidu to name a few. But none of them currently have the operating know-how of a small Pittsburgh-based company called Seegrid.

Seegrid is a company founded by Dr. Hans Moravec that specializes in VGV’s or Vison Guided Vehicles. These vision guided vehicles (pallet trucks) rely on a system called evidence grid technology which Dr. Moravec invented during his time as a research professor at Carnegie Mellon.  

The evidence grid technology is comprised of an array of “stereo cameras” that are designed and built in-house. Up to five pairs of cameras facing various directions map every square foot of their environment. What's interesting is that they don’t store all of the information because that amount of data would be overwhelming to the computers.   

                                                            

Evidence Grid Technology remembers its environment using specific points.
                                   

Instead, what they do is use Evidence Grids, which store only the important bits of data. Seegrid’s very own Chief Technology Officer Mitchell Weiss explains it best,

“Now, when you capture the images and identify the features, you remember their locations not as distances from you, but as which cube they fit in. You step forward a few feet, look around again, and add another check mark to each cube in which you identify a feature. Now it doesn’t matter how far you move, you only have to store the information about each cube, not a full set of data for each step you take”.

The software simultaneously compares what it sees to known data so even if the environment changes, it will remember where it is. This system makes it much easier for an autonomous vehicle to navigate and “remember" its surroundings. Currently, Seegrid markets its autonomous pallet trucks as being ready to operate in a new environment with a few minutes of setup time. All that is required for setup is to show the vehicle around its environment which it will then start to map which Seegrid refers to as WalkThroughThenWork. 

 

                                                            

An example of how the Seegrid system sees its environment

It seems easy, safe, and reliable enough to integrate onto pallet trucks used in warehouses, but could it be adopted by the vehicles we drive every day? Let’s look at the evidence.   

Currently, there are autonomous systems in development that use a combination of radar, cameras, and lasers among other things depending on who is developing the technology. These have all found varying levels of success, but Seegrid’s VGV’s are already in the field logging countless hours with a much simpler system.

Secondly, some modern vehicles already have cameras built-in for 360-degree views, integrating a few more wouldn’t take up any more room or drastically change the electronic package of the vehicle. Any hard drives or computers could find space where existing systems are packaged like behind the dashboard, under seats or rear compartment areas.

Lastly, vehicles wouldn’t need any special programming to “learn” the route that you take to work or to the grocery store. By simply driving your most common routes, the software would log that data and thereby be programmed to drive autonomously after that.  

 

                                     

Autonomous pallet trucks with Seegrid's VGV's

Of course, the government would probably like to see some testing and data for safety purposes, but the potential evidence for integrating Seegrid’s technology is compelling. Their vehicles already have a good track record of working in dangerous and constantly changing environments. Newer vehicles already have multi-camera systems and extremely powerful computers, so integration into existing vehicle systems could be easier. Programming these vehicles is potentially as easy as just driving them once. With Seegrid’s expertise in autonomy, I think they could prove to be a serious contender, if only they entered the ring.

 

Images courtesy of Seegrid

1 Comment
  • David Nobles April 15, 2016

    This is an interesting concept, and makes sense in the context of a warehouse.  But it seems that this system assumes an unchanging world.  Once the robot has seen the world, it knows where everything is.  What if other drivers are introduced to the problem?  How could this system guess the intentions are track the movement of other drivers?  This type of autonomous driving does not seem to address this issue at all.

    Like. Reply