We just got back from our first demo on a 150MW solar construction site, where we showed off our initial prototype: an autonomous forklift that unloads pallets of solar modules from a truck and stages them around the site. It’s a huge milestone for us, and we felt like now would be a good time to share what we’re working on more publicly. You can see a couple videos of our robot in action here:
Staging modules on the site: https://youtu.be/Fwf4v8upuoI
Performing a two-pallet sliding and unloading operation in our warehouse: https://youtu.be/EOJiyMXpVeQ
As solar modules have become commoditized and prices have plummeted, solar has become the cheapest form of power generation in many regions. Demand has skyrocketed, and now the primary barrier to getting it installed is labor logistics and bandwidth. Every solar construction company we've talked to is drowning in demand and turning down projects because they don't have the capacity to build them. 1/5th of all the solar that exists in the US was installed last year!
We're engineers who have been friends since living together at MIT where we studied robotics and CS. We always wanted to start a company together. We zeroed in on solar after seeing compelling statistics about its cost effectiveness and projected growth – and because we shared a motivation to do something about climate change. We actually started out writing software to predict optimal locations for solar sites (searching land for sale and scoring by price, amount of sunlight, proximity to existing substations) when we decided to learn more about what comes next.
Utility-scale solar farms (2MW+) are mechanically quite simple. They feature a steel racking system held to the ground by vertical posts ("piles"), and overwhelmingly (90%+) feature a single motorized axis to track the sun over the course of the day. Modules are then fastened to this axis with brackets.
We're using a two-part robotic system to build this racking structure. First, a portable robotic factory placed on-site assembles sections of racking hardware and solar modules. This factory fits inside a shipping container. Robotic arms pick up solar modules from a stack and fasten them to a long metal tube (the "torque tube"). Second, autonomous delivery vehicles distribute these assembled sections into the field and fasten them in place onto target destination piles.
This is a hard technical problem, but not research-level hard. We think of it as the "homework version" of self-driving cars, as we're operating in a semi-structured environment (flattened dirt field) with drastically fewer edge cases. Manual construction today breaks about 0.1%-0.5% of modules during installation, which is an easier bar for us to target than the stringent performance requirements of the AV world.
We're operating in a risk-averse industry, though, which makes deploying new technology more challenging. One industry-standard term we've become very familiar with is "bankability". It's difficult for projects to secure funding from lenders if they aren't using parts that have already spent years out in the field.
We've seen surprisingly little penetration of technology into this space in general. Projects are largely tracked with sticky notes in a "command room", material delivery schedules are highly volatile and often not known until days in advance, and there's no live monitoring of construction progress, making current status opaque. We actually had a site we visited outright lose a forklift – we were surprised that all vehicles aren't GPS tagged and monitored, especially given they're operating on multi-thousand acre sites.
Our system is the first to handle the full mechanical installation of existing solar components (remember bankability). We've tweaked the order of construction operations slightly to be more robot-friendly, as the more precise operations involved in fastening modules to steel tubes happen in a more controlled factory environment.
For our mobile robots, we’re building on top of existing vehicles (called telehandlers, or reach lifts) which are already ubiquitous on these sites due to their enormous tires and broad capabilities. They're able to unload shipping containers due to their extending boom, as well as move materials around the site. On our prototype vehicle, we did some significant up-front reverse engineering including mapping out CAN messages sans documentation. The steering and brake were directly hydraulically actuated (no drive-by-wire), so we added motors to both in order to control them with our software stack. The most unique sensor we contributed was an optical mouse sensor mounted onto the boom joint, telling us the extension distance.
The backbone of our robotic sensing is a robust vision system. We're using stereo cameras for SLAM and object detection. Fortunately, solar construction sites already have detailed engineering drawings including GPS coordinates of each vertical post in the ground, so we have a detailed map of the site to localize ourselves on.
Watching the existing process for large-scale solar installation in real time evokes the sense of watching paint dry or grass grow, only it involves hundreds of workers. After witnessing the physically grueling and inefficient process of workers manually installing thousands of solar modules, we realized there had to be a better way of building solar, and that increased automation was the way forward.
Our goal is to transition the world to renewable energy as quickly as possible. We’re excited to share what we’re working on with HN - please let us know what you think in the comments and we’ll be around to respond!
P.S. We’re hiring! If you want to work on cool robots with a positive climate impact, please reach out: https://chargerobotics.com/careers.html
by Andrey
Azimov