Autonomous vehicle R&D accelerates adoption of AI-based vegetation management
My previous blog on vegetation management finished with a teaser, regarding the falling costs of data collection, improving data science in computer vision. Here’s the follow-up, which starts off with the stalled progress in developing self-driving cars. I believe automotive’s loss will lead to utilities’ gain. Despite my innate conservatism, I’ll put my neck on the line and state categorically that the automotive industry will save network utilities billions in veg management costs over the next 10-15 years.
Self-driving vehicles are still a long way off.
Level 4 or 5 vehicle autonomy—where humans have little or no control over the vehicle—is proving to be a difficult nut to crack and is likely decades away from deployment (if it’s possible at all). Investors that have ploughed billions into autonomy R&D must quickly expand their focus to other industries, to get a financial return. Vegetation management is a powerful use case for two primary technologies used by autonomous vehicles: LiDAR sensing and machine vision.
LiDAR data collection has been an expensive exercise for utilities.
LiDAR (light detection and ranging) data is used to create an accurate 3D representation of the geography surrounding its network assets. To date, LiDAR has not been the cheapest data to collect. Until recently, LiDAR was a niche technology, and sensors were expensive. Also, the only way to record LiDAR data along a power line was by chartering a plane, or more recently to fly a drone. It should not come as a surprise, therefore, to learn that in the utility industry, LiDAR data collection is an annual process that is often restricted to certain parts of a service territory.
Expect LiDAR sensor manufacturers to aggressively target new markets, including utilities.
When the vehicle autonomy goldrush was in full swing, millions of dollars were ploughed into LiDAR—a foundation technology in advanced driver assist and autonomous vehicle systems—research and development. Consequently, LiDAR sensor costs have plummeted a hundred-fold, from tens of thousands of dollars to less than a hundred (for the most basic models).
Such low sensor costs could radically change the way utilities collect LiDAR data. It may make more economic sense to fix LiDAR sensors on poles and pylons, obviating the need to charter planes or fly drones. The data could be collected using existing network telemetry, so that an annual survey could be replaced by real-time imaging of network assets, with a relatively short payback period.
Machine vision innovation is intensifying; and veg management will benefit.
The vehicle autonomy arms race benefits AI-based veg management in other ways. Eye-watering amounts have been spent on machine vision and the complementary AI fields of deep learning and neural networks.
Machine vision is widely used in industrial automation, robots and drones. It can be used to augment existing security CCTV, manufacturing, logistics, and, of course, vehicle automation. All this attention creates another arms race, with VC-backed vendors eager to develop the next breakthrough in machine vision-based data science. Utilities—and veg management in particular—stand to benefit from the increased sophistication of machine vision, when analytics specialists look for new markets for their software.
AI-based veg management is a case of when, not if.
If power line vegetation management were the only application for LiDAR and machine vision, sensors would still cost tens of thousands, and machine vision algos would be coded by in-house IT or boutique service companies. Thankfully, it’s not. The vehicle autonomy arms race will benefit the utility industry, and it could be that in a few years veg management will be largely be underpinned by AI technology. In the meantime, my recommendation is for utilities to roll their sleeves up and get comfortable with the technology currently available, it’s only going to improve.