Soft robots, Part 2 — implications
What will soft, mobile robots be able to do that previous generations cannot? A lot. But I’m particularly intrigued by two large categories:
- Inspection, maintenance and repair.
- Health care/family care assistance.
There are still many things that are hard for humans to keep in good working order, including:
- Power lines.
- Anything that’s underwater (cables, drilling platforms, etc.)
- Pipelines, ducts, and water mains (especially from the inside).
- Any kind of geographically remote power station or other installation.
Sometimes the issue is (hopefully minor) repairs. Sometimes it’s cleaning or lubrication. In some cases one might want to upgrade a structure with fixed sensors, and the “repair” is mainly putting those sensors in place. In all these cases, it seems that soft robots could eventually offer a solution. Further examples, I’m sure, could be found in factories, mines, or farms.
Of course, if there’s a maintenance/repair need, inspection is at least part of the challenge; in some cases it’s almost the whole thing. And so this technology will help lead us toward the point that substantially all major objects will be associated with consistent flows of data. Opportunities for data analysis will abound.
One other point about data flows — suppose you have two kinds of machines that can do a task, one of which is flexible, the other rigid. The flexible one will naturally have much more variance in what happens from one instance of the task to the next one. That’s just another way in which soft robots will induce greater quantities of machine-generated data.
Let’s now consider health care, whose basic characteristics include:
- It’s done to people …
- … especially ones who don’t feel very good.
People who are sick, elderly or whatever can often use help with simple tasks — e.g., taking themselves to the bathroom, or fetching a glass water. So can their caretakers — e.g., turning a patient over in bed. That’s even before we get to more medical tasks such as checking and re-bandaging an awkwardly-placed wound. And on the healthier side, I wouldn’t mind having a robot around the house that could, for example, spot me with free weights. Fully general forms of this seem rather futuristic. But even limited forms might augment skilled-nurse labor, or let people stay in their own homes who at the moment can’t quite make it there.
And, once again, any of these use cases would likely be associated with its own stream(s) of observational and introspective data.
Related link
- Part 1 of this series was a quick introduction to soft and mobile robotics.
Comments
5 Responses to “Soft robots, Part 2 — implications”
Leave a Reply
[…] Now let’s turn to some of the implications of soft and mobile robotic technology. […]
Robotic repairs are not ready yet but autonomous robotic inspection of infrastructure is a nascent, high-growth market. As you correctly point out, the big advantage is that autonomous systems can (relatively) inexpensively go many places that are difficult for humans to get to, which also makes it possible to do inspections more thoroughly and frequently. (SpaceCurve does quite a bit of work in this area).
From a data platform perspective, there are a few thematic challenges:
– The sensor suites on autonomous robotic inspection systems throw off considerable amounts of sensor data, tens of TB per day is typical for a modest app. Many sources are real-time and continuous. There is some data reduction but it is still high volume and it needs to be retained. This would not be a big deal except that the workload is far outside the comfort zone of most big data platforms at this scale (see below).
– Analysis is built around modeling changes over time and being able to do exploratory drill down on historical sensor data to understand and classify anomalies, sometimes without a human in the loop. In practice, this means you need a long baseline of sensor data (not summarizations) that is online and quickly accessible. This fits neither the offline batch processing nor stream processing model of big data platforms. It is much more of a traditional online enterprise database workload, just on extremely large warm (but not hot) sensor data sets.
– For most types of autonomous inspection robots, the analysis is intrinsically spatial, and fairly sophisticated spatial analysis at that e.g. spatiotemporal joins over paths and polygonal features from different types of sources with high precision. This is by far the biggest capability gap for implementing these types of robotic sensor platforms; the Hadoop ecosystem does not offer it at all, with nothing on the horizon. You can often make compromises around data volume and velocity but not the ability to express the fundamental analysis.
Robotic sensor platforms, especially for infrastructure inspection, are moving forward about as fast as the technology development will allow because the value is obvious. Limited availability of spatial analytics suitable for sensor analysis, and to a lesser extent platforms that are optimized for the workload, have been among the major bottlenecks to progress. But give it a few years and it will be ubiquitous.
I think last couple of posts ( Innovation, Robots .. ) clearly show where future might be. It is also all driven by data – Google autonomous cars ( they are robots too ), ML all rely on fast processing of huge volumes of data. Additional benefit is actually seeing accurate physical movement as result of data crunching. Implications on physical world might be amazing. I know AI as an umbrella term with many applications disappointed in the past, but this time it could really be different, don’t you think ? And if so, then why ? Moore’s law can’t be the only explanation.
Last couple of posts (back to Graph analytics and data management still confused) indicates innovation of new appliance architecture (perhaps on-premises) to covert sensor suites into analog signals for merging together, then back into digital signals for in-cloud analytics and anomaly detection. Is the real AI singularity of volume, velocity, variety and variability by super low cost signal processing?
[…] The vulnerable (elderly, hospital patients) will be more accurately and comprehensively monitored. Also, their care will likely be aided by robotics. […]