Huawei RoadReacher underwhelms in MWC AI self-driving demo

One of the superficially most eye-catching demos at Mobile World Congress (MWC) was Huawei’s application of its Mate 10 Pro smartphone to drive a Porsche Panamera adapted for robotic operation. But closer inspection revealed that this was really just an attempt to galvanize developers into producing apps that exploit the neural processing unit (NPU) embedded in the Kirin 970 chip running the phone. To date Huawei has struggled to win interest in NPU from developers despite the well hyped AI tag, while the phone itself has failed to attract partnerships with any US carrier and therefore gain traction in that market.

The phone has though sold well in Europe, and MWC 2018 in Barcelona seemed the ideal stage to amplify interest in the NPU from both consumers and developers. But it was oversold by Huawei Western Europe CMO Andrew Garrihy’s statement that the company had succeeded in just 5 weeks in not only training the smart phone to drive a car but to “use its AI capabilities to see certain objects, and be taught to avoid them”. Surely that is what all driverless cars should be able to do, even at SAE Level 2 never mind Level 4, which enables autonomous driving under almost all circumstances.

The car was able to identify a dog and tell it apart from a cat, as well as take evasive action. In truth this was something of a gimmick, for it seems pretty clear that a smartphone lacks the resources to perform all tasks needed for full blown autonomous driving. This was just to show the image processing capabilities of the smart phone’s NPU and that succeeded as far as it went, but highlighted the challenge of attracting interest. Until recently at least, there was just one smartphone app apart from Huawei’s own camera software using the NPU and that was Microsoft’s Translator.

The camera software itself perhaps highlighted why it has been hard to attract interest, because it is not clear what the AI part adds to the already very good camera hardware, in terms of quality. The idea is that by identifying objects, it can tweak the exposure time and focus more accurately to reflect the nature of the composition, but the benefits might only really be worth having on a dedicated professional camera.

It may well be that autonomous cars will benefit from integration of such capabilities to fine-tune their responses to given situations. It might decide that it is worth taking more risks to avoid running over a dog than a cat for example, although not all drivers would agree.

Otherwise, there was a lack of high profile connected car demos at MWC – far less than the more consumer focused CES 2018 in January. This reflects the nature of the show, but also the state of play in the industry, which is a relatively slow burn entailing a lot of hard graft integrating different players within a complex ecosystem.

This was indicated by one announcement at MWC of a partnership between London-based analytics and cloud services developer UST Global and Spanish IoT platform provider Net4Things. This project is aimed at mobile carriers in six specific Latin countries, Spain, Portugal, Brazil, Chile, Colombia and Peru, integrating other aspects of an end-to-end car ecosystem including navigation and insurance.

The logic of this choice of countries may be that Spain and Portugal share a common border, while the other four are contiguous. The partnership could therefore help mobile operators in these regions develop a package they can take to automobile OEMs in the region supporting cross border connected car operation – perhaps an easier sale than a new AI chip from a company currently beset by (perhaps unwarranted) US security scrutiny.