From cleaning robots to self-driving cars, simulation may be the only way to train autonomous vehicles.
In the early days of the COVID-19 pandemic, it became clear that the current industrial cleaning solutions would not be sufficient. Faced with both staffing shortages and a need for enhanced cleaning procedures, most companies knew it would be difficult to continue ensuring the health and safety of their workplaces. Knowing that autonomous cleaners like the Roomba vacuum were already successful in consumer homes, some companies aimed to utilize autonomous solutions to meet their cleaning needs. To help address this increased demand for industrial cleaners, Avidbots launched Neo 2, a fully autonomous robot floor cleaner, in 2020.
Last month, Avidbots announced a partnership with software developer Maplesoft to improve its autonomous cleaning robots. Avidbots will use MapleSim, Maplesoft’s system-level modeling and simulation tool, to develop and test its virtual prototypes and train its AI in simulated environments. In so doing, Avidbots hopes to save the time and money it would typically require to build physical prototypes and manually test and train them in diverse spaces.
“Maplesoft’s advanced virtual modeling capabilities help to keep our product development nimble and adaptive, so we can deliver our customers built-in intelligence that allows Neo to perform on its own with a consistent, efficient, and measurable clean,” said Peter Ahn, VP of Engineering at Avidbots, in a news release announcing the collaboration.
Neo 2 is an Autonomous Cleaning Robot for Industrial Applications
Avidbots describes Neo as a robot designed to clean more effectively and efficiently than humans. The company claims that Neo can autonomously clean 1.35 meters of space per second and 3,900 m2 of space per hour.
The speed and utility of Neo are driven by its ability to detect and adapt to obstacles in real time. So, how does it work?
It starts with a fully customized cleaning plan that is developed for each new deployment of a robot. Then, each time Neo begins cleaning, it compares the current environment to the original plan to maximize the disinfection based on current obstacles. Avidbots’ proprietary AI-based solution is trained to identify and avoid obstacles while maximizing the disinfection area. Using long-range LiDAR sensor data and the original input cleaning maps, Neo is given a location-level spatial awareness of its whereabouts relative to items, people, walls and more. The level of awareness that is possible with the proprietary AI allows faster decision-making and real-time responses to the ever-changing environment of industrial buildings.
This differs significantly from other autonomous cleaning technologies, says Avidbots, which claims that other products are more accurately described as semiautonomous. Most of these technologies rely on the traditional teach-and-repeat principle, where the robot relies on a set of possible input routes. This means that if the robot faces an environmental change, it will stop cleaning and require a manual override to continue on its path. Although the robot will learn from these situations, the result is a technology that requires significant human intervention. Plus, the actual cleaning process is interrupted and may be less effective.
The Neo cleaners also include the Avidbots Command Center, which generates in-depth performance reports, real-time monitoring and metrics, and management of multiple cleaning robots to allow for higher-level decision-making. Companies can glean the time spent cleaning, the total area cleaned, and the amount of water used in detail to ensure that their workspaces are safe and in line with current KPIs.
Maplesoft Provides Simulation Environment to Accelerate Product Development
Autonomous technology is always in a constant state of improvement. But many people continue to wonder why fully autonomous vehicles remain elusive. Although several issues are holding back the full realization of the technology, one of the main challenges is predicting the inherently chaotic natural environment. Beyond vehicles, predicting the unpredictable also remains a significant bottleneck for other autonomous technologies.
Autonomous vehicles and other robots rely on extensive testing to train robots for the many possible scenarios they will encounter in the real world. Usually, this means a company needs to build successive prototypes and test them in diverse environments. However, this strategy is time-consuming, expensive, and prohibitive to innovation.
That’s why AI companies like Avidbots are turning to simulation software to accelerate their early R&D pipelines.
“Maplesoft has always provided advanced modeling tools that equip tech companies to design better, stronger machines through simulation,” said Chris Harduwar, VP of Business Development at Maplesoft, in the press release.
Although it remains unclear exactly how Avidbots is using MapleSim to design and develop its models, the software has been used to develop other autonomous technologies in the past. The Motion Research Group at the University of Waterloo’s Centre for Automotive Research used MapleSim to support its autonomous vehicle development. The team created high-fidelity models of their vehicle in MapleSim and validated them using track testing, four-post testing, and rolling dyno tests. The final simulation code was then exported and tested in virtual reality environments created using Unreal Engine 4.
Simulation Software Is Driving Autonomous Development
For more than a decade, the automotive industry has been focused on developing better simulation solutions to support the R&D required to bring autonomous vehicles into our everyday lives. In 2016, Ansys determined that simulation solutions would save companies thousands of years of physical testing during the development of autonomous technologies. Now, Ansys offers a specialized solution for autonomous vehicles and advanced driver assistance systems simulation. Late last year, we shared a success story from the partnership between Ford and Ansys on an advanced vehicle lighting system developed in part with simulation software.
Simulation testing helps companies not only reach their required test mileage, but it can also help to save lives by making edge case testing easier before physical prototypes are put on the road. Knowing the importance of these solutions, Ansys is not alone when it comes to simulation software designed for autonomous vehicles. For example, Claytex has been working on simulation software specifically for autonomous vehicles since 2016. In 2017, Siemens Digital Factory acquired TASS International, another simulation developer aimed at autonomous vehicle applications, to expand its simulation capabilities. Companies like Cognata are also looking to help engineers meet their test mileage with detailed AR environments and a physics and deep learning-driven simulation engine.
With Neo already used in airports, warehouses and retail facilities, the need for autonomous cleaning solutions is clearly expanding. The partnership of Maplesoft and Avidbots is building on years of investment by autonomous vehicle companies in simulation software and development.