NVIDIA GPU Conference – Boosting Drone & Robot Performance


The NVIDIA GPU Conference, which we attended from April 4-7, gained considerable publicity because of the Woz (Steve Wozniak) marveling at NVIDIA’s virtual reality visit to Mars. However, of special interest to us, and many others, was the Emerging Companies Summit. Here 12 new companies, using GPU (graphics processing unit) technology from the host, squared off before a panel of venture investor experts.

The poster child past winner of this event, by the way, was Oculus. We divided this year’s presenters, perhaps arbitrarily, into two main categories:

1. Dealing with Drones or Robotics

2. Enhancing A Range Of Other Experiences

In this article we discuss four presenters that relate to Category 1, Drones or Robotics.

Making Drones More Resilient

Autonomous Drones
An Israeli-based company Aerial Guard showed a video of a self-directed drone that was programmed to follow a bicycle rider. First the rider went down an open path, but then entered a heavily wooded area. The drone followed, careful to avoid the trees.

Finally, the drone encountered a tight thicket of trees making the path ahead unpassable. It backed off, circled a bit and then calculated a new route to catch up to its prey.

Aerial Guard “provides autonomous situational awareness for drones and UAVs.” This is designed to increase safety, survivability, and mission capabilities.

All-Weather Drones
A separate issue – drone flight in bad weather – is addressed by China-based BriSky Technology. BriSky is aiming at the industrial drone market. It claims that by use of computer vision and deep learning it can offer “inspections of power lines and wind turbines, traffic monitoring, surveillance, and public security” even in harsh weather conditions.

It should be noted that all of the participants in the program are users of NVIDIA’s GPU high performance computing (HPC) capabilities. Therefore, when asked by the panel of experts about how much weight their componentry added to the drone payload, the answers were predictably – not much, because of the use of the host company’s HPC capabilities.

Drone-Related Vision Capability

Entropix, a U.S.-based company, is not strictly a drone-oriented provider. However, it does feature the fact that it enables cameras, “like those in smartphones and drones” to capture extremely high resolution images.

It states that through software processing capability it brings “high resolution video content creation into the realm of computational imaging,” claiming 8K resolution. (8K has 16 times the number of pixels of Full HD – 1080p – and four times the number of pixels of 4K, which is driving current TV sales growth.)

The magic sauce is provided in Entropix’ GPU powered post-processing cloud engine. The CEO was even joshed by panel members for his repeated credits to NVIDIA and its GPU technology.

And The Winner Is … The Robot Recycler

The winner of the competition was Sadako Technologies, a company based in Spain that has built a robotic device that combines machine learning/computer vision – to sort through and identify objects in recycled waste dumps – with a specialized gripping device that can extract specific items, for example, bottles.

The robot basically sits above a streaming pile of waste and can grasp and extract one item every three seconds. The company points out that using current, manual, methods, only 13% of 1.9 billion tons of waste every year ever gets recycled. The robot can greatly increase the level of recycled material. The company was presented with a $100,000 check for its selection as the most interesting, from a commercial perspective, by the experts and audience vote.

Our Take

Sadako’s vision is interesting because it is using deep learning, enabled by GPU-based HPC on the cloud and artificial intelligence to instruct machines to be able to identify very complex objects. Thus they can distinguish specific types of items in a moving stream of waste.

The company states that in some cases, the robots’ ability to identify items exceeds the recognition capability of humans. They foresee that this line of R&D will enable them to create machines to address other complex tasks, which would represent an advance over the current predominant use of robots for repetitive operations

Autonomous Drones
The logic in favor of autonomous drones is obvious; i.e., if we’re going to have autonomous road-based vehicles, why not autonomous UAVs? DARPA (Defense Advanced Research Projects Agency), as well as students at the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and others have been working on autonomous drones. (See, e.g., “New algorithm allows autonomous drone to zip through trees at 30 mph,” gizmag.com 11/9/15.)

Issues are being raised about use of autonomous drones by the military, particularly in attack mode. (“Navy seeks autonomous drones despite warnings from critics,” cnn.com, 2/16/16.) Aerial Guard suggested in their presentation that they had experience working with the Israeli military. Regardless of the military issues, it would appear that autonomous drones are likely to become widely available, at least for industrial and consumer use.

All-Weather Drones
The market for all-weather drones is in existence and growing. For example, Aeryon Labs, which we wrote about recently highlights the use of its SkyRanger drone in very poor weather conditions. (“Aeryon Labs – The Surging Growth of Drones,” mobilecloudera.com, 12/13/16). Others, such as Trimble Navigation, also offer all-weather UAVs.

Regarding BriSky, it would appear that the company’s attractiveness may well depend primarily on its ability to develop compelling industrial apps through the use of computer vision and deep learning, rather than the all-weather fitness of its drones, since the latter appears to be a fairly competitive area. Interestingly, it is this same combination of technologies that underlies the success of Sadako’s approach to robots.

Drone-Related Vision Capability
When we first became involved with 4K, about three-to-four years ago, it appeared to face a difficult challenge to gain market acceptance. Now 4K TV sales are soaring, up nearly 200% in 2015.

However, there already is a race to move on to 8K (even 11K!) for TVs, smartphone cameras and drone cameras. Entropix appears to be offering a software solution that can serve the embedded base of earlier vintage cameras in smartphones and drones. As respects cameras in drones, the issue is how long the earlier generation cameras will remain in widespread use, which may be more short-lived than we might have expected even a year or two ago.

Other Emerging Company Contestants

In the second category in the Emerging Company Summit were seven companies that dealt with enhancing experiences in a wide range of categories.:

+ Facial Recognition – Linkface
+ 3D Cameras For Virtual Reality – Lucid
+ Food Flavors – Analytical Flavor Systems
+ Weather Forecasting – TempoQuest
+ Speech-To-Text – Intelligent Voice
+ Assistance For The Visually Impaired – Horus Technology
+ Customer Inquiries – Cognicor

Finally, there was one company that intends to launch satellites with hyper spectral capability that can provide incredibly detailed analysis of chemical composition of the earth’s surface, Hyper Cubes.

We plan coverage of some of these companies in the future.

Visit NVIDIA Website: www.nvidia.com/object/what-is-gpu-computing