I wrote about the Velodyne sensor a couple of weeks ago touting its commercial viability in the emerging mobile robotics market. I had a chance to talk to one of its inventors today. Velodyne President Bruce Hall made it clear that although his company doesn’t have a vehicular entrant, they are clearly and very visibly participating in the DARPA Urban Challenge.
“We went from being on 12 out of 35 teams at the beginning of the National Qualifying Event to being on 7 out of 11 at the finals. I hope that when vehicles 1, 2, and 3 come in, we’re on 3 out of 3 vehicles,” Hall said.
The Urban Challenge teams are comprised of some of the most brilliant engineers on the planet. Adoption of a technology that can give them a structured view of the world around their robot obviously seems too compelling to resist, considering the adoption rates of Velodyne’s LIDAR within the DARPA Urban Challenge ecosystem. The interpretation of the data is where the scientists shine. Mere mortal application developers, however, require a higher-level abstraction that interprets the data for them–recognizing a blob of points as a vehicle or a pedestrian, for example.
Every team that uses the Velodyne system has built what is likely to be a driver intended to solve the same problem. That seems to be the point. One of those implementations will be the best. “It’s a competition. These guys are trying to solve a common problem better than the rest,” Hall explained. That statement seems to be validated by several teams with whom I spoke. Teams that used a competing technology, ibeo, chose to employ their own object recognition algorithms instead of using the algorithms provided by ibeo, a technology I will cover in a subsequent post.
“Today, our customers require a certain level of sophistication that inherently limits our sales [to scientific programmers that can interpret the data]. We know that. Some customers are asking for (inerntial management) correction, object identification, and so on. Like any good business, we try to stay focused.”
Staying focused is not the same thing as ignoring customer feedback. “Customers wanted to be able to choose density over speed. We gave them the ability to spin the unit at 300 and at 900 RPM.” The slower-spinning mode provides higher density of data points while the higher speeds senses the world faster. Another feature due to customer feedback was the governing of the high data volume being broadcast by the device. The governor reduces the number of points sent by the device to something relatively consumable by most applications. The device is capable of informing the robot of over 2.3 million points every second but for the software that consumes it but it’s like drinking from a full-open fire hose. The device can be governed down to a slower 1 million points per second and the application can choose to ignore data it can’t consume.
In all emerging technologies there are always issues that come up that allow their makers to refine and to strengthen the technology. Velodyne is not immune to this. There is a little bit of tension in the air with regards to a bug that has manifested in the context of the Urban Challenge. Close-proximity RF signals can potentially confuse the sensor. This problem has a simple fix that was offered to all teams. All but one chose to decline the fix. Hendrik Dahlkamp of Stanford Univeristy reflected a classic product development discipline. “We declined on putting in the fix because it’s too risky to make a change this late in the game. Plus, the cost of the bug isn’t that great to us.” For another team, however, it’s life and death. Team Annie Way was almost eliminated when the Velodyne sensor was “blinded” by the radio of a DARPA official. Actually, the team was in fact eliminated until the Team Leader contested the decision and explained the situation. Team Annie Way is now a finalist and Velodyne is working very closely with them.