Combining Machine Learning and 5G in Robotics

With the advent of 5G connectivity and Machine Learning (ML) processes on ever smaller devices, it was inevitable that robotics would eventually benefit from both. Thus, it is unsurprising that a big player such as Qualcomm would come up with a platform for the purpose of making ML and 5G available to roboticists. Recently launched, the RB6 platform, along with the RB5 Autonomous Mobile Robot (AMR) Reference Design, come with very ambitious goals to further enhance the autonomy of autonomous mobile robots. Aimed at helping developers create intelligent robots, ‘intelligent’ in this case meaning the possibility to carry out ML processes on the robots themselves, the RB6 Platform comes with a suite of advanced tools for this purpose, from 5G connectivity supporting global sub-6GHz and millimeter Wave bands to the edge ML and video processing capabilities supported by a Qualcomm AI Engine capable of carrying out 70-200 trillion operations per second. The RB5 AMR Reference Design, in turn, offers a technical blueprint of an AMR with tightly integrated ML and 5G capabilities. This is a welcomed addition to Qualcomm’s robotics reference design lineup that includes Qualcomm Flight RB5 Platform. An all-in-one solution, the RB6 Platform and RB5 Reference Design together provide the necessary hardware and software development tools to make robots smarter, faster, and safer for everyone.

Such a set-up is obviously rather attractive, but there are also some significant shortcomings. On-device ML processes are typically of the inferential kind, as little-to-no training is conducted at the edge at present. Thus, this will require the requisite training to be conducted on a server elsewhere, with the result of such training then to be applied as an inference in whatever setting a robot is placed. This introduces unavoidable delays and perhaps not as much autonomy as needed for certain applications—last-mile delivery is one of the targets of Qualcomm’s Platform and Reference Design combo, for instance—not to mention the resources required to conduct training-and-then-inference processes. On-device training is a distant goal at present, at least for the more advanced processes, and while the possibility of applying inferences can clearly augment a robot’s capabilities, the real benefits will be reaped when a robot is capable of applying training models on data that itself has processed in situ.

5G connectivity, for its part, while offering greater and higher bandwidth than 4G and Wi-Fi networks, meaning a faster and wider connection overall, will at present work better outdoors than indoors in the case of robots, especially in the case of millimeter Wave bands, which have reduced range and penetrability compared to other cellular networks. This is particularly an issue in manufacturing and within warehouses, where the amount of shelving and metal can interfere with high-band frequencies between 24 and 47 GHz, though in sub-6 GHz frequencies 5G signals can work as well as WiFi-6 in indoors settings. Since this will require many more cells for a smooth functioning, this kind of set-up may well affect its application in robotics, as in the already-mentioned last-mile delivery. A faster and broader network obviously offers great advantages, but the roll-out and implementation of 5G networks remains a goal to be attained, in industry and elsewhere.

The robotics market is benefitting a great deal from endeavors such as Qualcomm’s, and ABI Research expects faster, more comprehensive platforms and reference designs to be available to vendors in the near future. This also includes hardware accelerators of various kinds, agnosticism regarding operating systems and software more generally, and simply more collaboration in developing robotic tools. This is where the market is headed, and some kind of integration of 5G connectivity as well as of ML processes within robots is the next natural step. However, it is also the case that these measures are, in a way, necessary patches over some of the very real deficiencies in the market. Indeed, in recent years there is a significant shortage of engineers in robotics and Artificial Intelligence (AI), the latter broadly construed, and these recent developments constitute a reflection of the general state of affairs.

While recognizing the importance of software-based tools in the development of robots, as this is what Qualcomm’s recently-launched products effectively are, ABI Research is of the opinion that more sophisticated development and research is being hampered by this shortage in expertise. As such, these new tools are likely to offer a short-term solution rather than something long-term and this ought to be of some concern, as in the long run there may be diminishing returns with this approach. ABI Research believes that there should be more investment in areas that require experts from different fields and that are moreover resource- and time-intensive, as is the case in robotics. Such investment would benefit from a collaboration with public research centers and institutions, which would serve the needs of industry more adequately.