Nvidia's Inception Awards spotlights innovative AI startups

Three winners were awarded $1 million in prizes Tuesday at Nvidia’s Inception Awards, an AI startup challenge. The finale was held at the company’s annual GPU Technology Conference in San Jose, and there was a “Shark Tank” vibe to the proceedings for the six finalist teams, whittled down from a dozen semi-finalists earlier in the month.

Four on-stage judges were given one minute to ask questions after each team’s five-minute presentation. The audience of about 650 was allowed to vote via an app, and the audience’s combined vote was averaged in as a fifth judge’s score, to determine winners.

The finalists gave an interesting look at how AI is poised to dramatically challenge industries in three areas: healthcare, enterprise and autonomous systems. Each area included two finalists, and the judging scores awarded one winner in each category. In healthcare, the winner was Subtle Medical; in enterprise, it was AiFi; and in autonomous systems, it was Kinema Systems.

Following is a look at all six finalists (with the three winners first), and what each is working on to attempt to positively change the future.

Subtle Medical (subtlemedical.com), a team of medical imaging researchers, AI pioneers and active entrepreneurs from Stanford, aims to use deep learning to greatly speed up the time it takes to perform MRIs-as well as make the resulting images higher resolution. The Deep Learning technologies applied will improve medical imaging for better acquisition, reconstruction, processing and analysis.

As a side benefit, the amount a radiation a patient is subjected to would decrease roughly by a factor of 200. This smaller amount of radiation was described as equivalent to the amount that many attendees experienced flying from New York City to the Bay Area. The more detailed MRI images may be able to even predict future medical issues.

AiFi (aifi.io) has a bit of an earth-shaking premise. It wants to make all stores checkout-less. The flexible and scalable model works for any existing stores, according to the company-meaning everything from small mom-and-pop convenience shops to major mega retailers comprising tens of thousands of square feet.

The company’s checkout-free technology uses AI algorithms real-time customer tracking,
action recognition, and product recognition. Its sophisticated camera technology can adapt to any type of store, and systems continuously track hundreds or thousands of shoppers in a store. It can recognize or re-identify them throughout a complete shopping session. The technology also enables a comprehensive understanding of shopping behaviors and gestures (even identifying
abnormal gestures) and identifies people who are shopping together as a group. It can recognize tens of thousands of SKU item numbers based on AI.

The solution does not require any major retrofitting on the part of the stores, and AiFi says that once installed, retailers will gain improved inventory management data and valuable insights
into consumer shopping habits and product preferences.

Kinema Systems (kinemasystems.com) is a Silicon Valley startup that has developed a system of deep learning 3D vision for industrial robots, allowing them to more effectively perform logistical problems, such as picking up and de-palletizing boxes. The team includes experts in robotics, automation and perception, with experience implementing and deploying innovative solutions to complex industrial problems.

The company’s initial product, Kinema Pick, is designed for de-palletization of multi-SKU, single SKU and random pallets. It can be integrated with any robot and vacuum gripper. Kinema Pick is easy to configure using a browser-based GUI. It integrates the KS1000, a high-resolution 3D/2D sensor with advanced motion planning and requires minimal training for any type of box or workcell, minimizing integration time.

The company says its solution is self-calibrating and deployable on standard or industrial PCs. It uses GPU-accelerated deep learning and 3D Vision to locate boxes, as well as perform label detection and alignment.

Cambridge Bio-Augmentation Systems (cbas.global) is attempting to create a common standard, a sort of USB connector for the body. The company calls this the Prosthetic Interface Device, or PID, and says it will increase the accessibility, functionality and effectiveness of bionic limbs. Its mission is to create the open standard platform to support high functionality neural and bionic treatments, making it easy for clinicians, engineers and researchers to develop and deliver superior treatments to patients.

The nervous system carries electrical impulses to control virtually every organ and bodily function. Many common chronic conditions often occur as a result of a failure or change in our neural pathways. Cambridge says if it can understand and correct these signals in real time, for each individual patient, the company can treat chronic illnesses in an effective, automated, and personalized way.

Cambridge says it has generated the world’s largest neural dataset, gathered from pre-clinical trials. The company is exploring how it can understand nerve signals in the limbs and interpret movements through machine learning and AI-and applying these methods to autonomic nervous signals to learn about general health patterns and abnormalities.

CrowdAI (crowdai.com) is a team of highly skilled engineers and scientists focused on creating up-to-date maps from satellite imagery using AI. The Geospatial analytics market is estimated at $8.2 billion and can involve everything from road maps for ride share companies to accessing building damage for insurance companies. As an example, the company’s technology found all the roads in Syria in six hours.

CrowdAI’s focuses include infrastructure (locating every road and building, mapping out road-level data), change detection (highlighting new construction, deforestation, or drought conditions), and scalable answers (up-to-date data at a country scale).

Ghost Robotics (ghostrobotics.io) has developed unmanned ground vehicles (UGVs) that the company says are superior to wheels and tracks when venturing outdoors. These quad-legged robotics feature reduced mechanical complexity with total software (SDK) control when compared to other legged and traditional wheeled and tracked UGVs on the market.

The company says that in reducing complexity, it inherently increases durability, agility and battery life, and reduces the cost to build and deploy autonomous robots. The modular design model allows ecosystem vendors to build solution specific quad-UGVs for virtually any use-case.

Ghost has a portfolio of easy-to-operate legged and hybrid UGVs from very small, foldable and ultra-fast ISR and public safety devices, including disposable warfare and marine swimmer models; to small and medium asset inspection, scientific and in-building security robots; and larger perimeter and all-campus security, logistics and delivery, and heavy task-mules. The can function as tele-operated, assisted or fully autonomous.

Nvidia
www.nvidia.com/page/home.html