Written by Gábor Pongrácz / Posted at 10/12/23
Enhancing Adaptability with Sensor Virtualization: How Our Perception Stack Seamlessly Handles New Fleets
In the rapidly evolving world of automated driving, adaptability is key. As our roadways become increasingly populated with vehicles of all shapes and sizes, seamlessly integrating automated driving systems in new fleets is paramount. At aiMotive, we've taken a giant leap in achieving this level of adaptability with our neural network technology inside our perception stack. In this blog post, we'll explore how our neural network effortlessly handles the transition from an older fleet of vehicles to a brand-new one while maintaining peak efficiency.
The Challenge of Change
Imagine this scenario: you've been using a fleet of classic sedans for your autonomous driving tests, and they've served you well. However, as your ambitions grow – or when Stellantis acquires your company – you decide to introduce a fleet of large SUVs into your testing environment. This transition brings with it a host of challenges.
Firstly, the new cars can be significantly larger, meaning they have different driving dynamics and sensor positions. Secondly, due to the increase in features, the sensor suite needs to be modified, adding new sensors to new locations – in our case moving from 6 cameras to a whopping 11. This shift can be a daunting prospect for most automated driving systems, as they often require extensive reconfiguration and retraining to accommodate these changes. But not ours.
Seamless Integration with aiDrive
Our neural network, which powers aiDrive, is designed with adaptability in mind. It's not bound by the constraints of specific vehicle types or sensor setups. Instead, it has the remarkable ability to be adjusted without additional effort or re-training. This means that when you introduce a new fleet of vehicles, no matter how different they are from your previous ones, the perception stack can seamlessly adapt without any extensive development or tinkering.
Plug and Play Efficiency with Virtual Sensors
One of the most astonishing aspects of this transition is the minimal impact it has on the overall efficiency of our automated driving system. You might expect that integrating a new fleet would require significant recalibration and fine-tuning, potentially leading to a drop in performance. However, our neural network defies these expectations due to the introduction of virtual sensors into the development process.
Thanks to its robust and smart design with such virtual sensors, the neural network can quickly understand the intricacies of the new vehicle setup and sensor positions. Adaptations take place offline in a few simple steps, ensuring that the aiDrive software stack continues to function optimally. The concept of virtual sensors allows not just for changing one camera to another or adapting its position, it also allows for bootstrapping new, previously unseen locations and orientations. This "plug and play" efficiency means that you can seamlessly switch between vehicle fleets without sacrificing the quality of your automated driving solution and not taking dents in development and feature rollout speed.
The Future of Fleet Adaptability
As the automotive industry continues to evolve, the ability to integrate new vehicle fleets effortlessly will become even more critical. Whether you're adding larger vehicles, different sensor configurations, or entirely new vehicle types to your autonomous testing environment, our neural network technology ensures that the transition is smooth, efficient, and hassle-free.
In conclusion, at aiMotive, we're not just building automated driving solutions; we're building adaptive solutions. Our neural network's remarkable ability to handle the transition from old fleets to new ones without additional development or external tinkering is a testament to our commitment to innovation and efficiency in the world of automated driving. As the road ahead unfolds, rest assured that our technology will continue to lead the way.