Welcome to the 5th installment of our Introduction to Autonomous Mowing series. This week we will look at our autonomous “Stack” (mobility stack, driving technology stack, and many terms for the same thing). Over the next couple of posts, all significant components and sensor packages will be reviewed. This understanding of the interactions will support your knowledge of our chosen sensor suite.
The autonomous stack is a term initially defined within the autonomous automobile industry. If we are to oversimplify things, think about your favorite breakfast pancake stack – multiple layers of carbs and calories stacked on each other – where pancakes are swapped out for multiple levels of hardware and software (I know – it is a stretch – but).
The autonomous driving technology stack contains five primary layers. Again, defined differently by many in the industry:
Sensors or hardware
Infrastructure and off-board software and data
Computing methodology and network
The product or “applications.”
All these layers together form the final application. The graph below pictorially represents the makeup of each group in the final product. Getting this software/hardware combination right is the key to success in the autonomous industry. The developed onboard software finalization can be challenging given the package must be user-friendly, require minimal training, have no additional environmental setup, and have robust reporting.
The second key to success in developing or modifying a vehicle with autonomous navigation is precise and accurate motion control. Our engineers have spent more time developing the steering algorithms and hardware than any other system on Nomad.
Critical stack enablers
Many areas cut across and enhance the stack, impacting internal functionality, external value addition, or constraint. For example, as seen in simulation and AI frameworks. Simulation is used to reduce the costs of live testing and validate against multiple “what if” scenarios. Simulation is necessary to ensure the stack improves and provides consistent results in the real world. AI frameworks are not part of the stack but are a critical component of how imaging systems recognize patterns, images, predict pedestrian and vehicle behaviors, etc. AI models resulting from training are an essential part of the decision flow in our system.
Security is an intrinsic part of the entire stack, with regulation and safety needs driving choices within the stack design. Some security elements are ensuring GNSS inputs are not compromised, validating that video content is coming from onboard cameras, and providing secure connectivity and isolation of networks.
Finally, we have designed this “Stack” to help facilitate simple installation for any new mowing site. We have reviewed our mapping and localization process in previous articles, but the “Stack” structure allows us to complete the installation process simply and quickly.
We look forward to your comments on our Introduction to Autonomous Mowing series. Look out for upcoming topics such as Nomad Specifications and Stories from the Field after we review the sensor package in our next post.