Manufacturers and logistics providers have an an immense and growing need for flexibility. Mobile robots with increasingly autonomous navigation and common interfaces can help meet this need, as new and maturing technologies take robotics to new levels of industrial utilization.
The Fraunhofer Institute for Manufacturing Engineering and Automation (IPA) in Stuttgart, Germany, has been developing its NODE technology to improve the navigation of autonomous mobile robots.
From AGVs to AMRs
Automated guided vehicles (AGVs) have been a major component of the recent expansion in commercial service robots. Almost 111,000 units were sold in 2018, an increase of 53% in sales and 60% in units compared with 2017, according to the International Federation of Robotics. Of these, almost 8,000 were involved in production, and the rest were primarily in the e-commerce sector.
While some mobile robot applications are still feasible with the rigid structures used by AGVs, such as physical tracks, many dynamic environments require more agile robots. The trend toward smaller batches and higher product variability requires greater flexibility in production and materials handling. Autonomous mobile robots (AMRs) use adaptive navigation algorithms to learn new routes and meet this need.
Concentration and mixed fleets require sophisticated software
Two additional trends are occurring in mobile robots. The first is concentration. As more robotic vehicles drive in an environment, software developers have responded with more efficient systems for fleet management, traffic control, and dynamic path planning.
The second trend is toward heterogeneous fleets. Many AMRs are equipped for specific processes, and large facilities may have multiple types of robots from different manufacturers. Many vehicles can communicate only with similar robots.
The has been progress here with VDA 5050, a new interface proposed by the German Association of the Automotive Industry. In future, this interface should become an international standard.
What robots need in autonomous navigation
As mobile robots move in more challenging environments and cooperate more among themselves and with other systems, both hardware and software must evolve. In its autonomous navigation research and development, Fraunhofer IPA identified the following requirements:
- Robots must work without infrastructure and markers. AMRs eliminate the costs and effort involved in installing and maintaining AGVs.
- Software should be easy to use, with intuitive user interfaces and algorithms for self-configuration and self-optimization. This enables users without expert knowledge to put new applications into operation in the space of just a few hours.
- Autonomous navigation software must be flexible. Thanks to their ability to adapt to changing environmental conditions, AMRs should be usable in a wide range of applications.
- A fleet should also easily be expandable to include virtual robots. With the help of augmented reality, travel paths and other information can be visualized. This simplifies and accelerates the commissioning, maintenance, and adjustments of the fleet.
Fraunhofer IPA develops NODE
Fraunhofer IPA has developed the Navigation on Demand, or NODE orchestration, coordination, and navigation system, to meet the requirements outlined above. It builds a common database by cross-linking vehicles, both among themselves and with external computing resources. Thanks to this common database, each vehicle always has access to the sensor data of the entire fleet.
The cooperative navigation algorithms use this database for optimal fleet control. Previously, it was possible to control the navigation of only one vehicle optimally according to its local field of view. Now, an entire fleet can be operated based on the aggregated knowledge.
By connecting to a cloud/edge infrastructure, computationally intensive processes can be outsourced to reduce cost-intensive local computing resources on the robots. Furthermore, it enables easy deployment and software updates, as well as remote monitoring and analysis of the robots.
Applying machine learning to autonomous navigation
Fraunhofer’s NODE uses machine-learning methods with the aim using the data collected by the fleet to improve mobile robot autonomy and efficiency. It can also reduce the manual set-up effort.
In this context, the NODE team is currently working on three challenges. The first is the experience-based optimization of global route planning. For this purpose, virtual vehicles are driven first to determine available routes. Then the data from real vehicles is used to adjust route costs based on operational data.
In the second topic, the navigation experts let the software learn in a simulated environment how to control a vehicle to follow a route and at the same time avoid both static and dynamic obstacles. This takes vehicle characteristics such as the chassis or necessary safety distances during different driving situations into account. With the help of reinforcement learning — i.e., reward-based learning — the team can develop strategies for solving specific traffic situations efficiently. The lessons are then transferred to real vehicles.
For the last autonomous navigation challenge, the NODE team is working on mutual detection and cooperative localization using machine-learning methods. As vehicles recognize each other and thus determine their relative position, localization will be more robust, and vehicles with less powerful sensors will benefit from sensors of other vehicles. This method is also helpful if sensor ranges are short and the environments are large or dynamic at the same time.
Different versions of this software have already been implemented in machines ranging from vacuum cleaning robots to self-driving trucks. Autonomous navigation techniques are in continuous and successful use in industrial operations, and improvements should widen robotics applications. More information and references for the automotive industry can be found at the NODE website.
About the author
Stefan Dörr is project manager within the Industrial and Commercial Service Robots team at Fraunhofer IPA. Contact him at firstname.lastname@example.org.
The post How to optimize autonomous navigation through networking appeared first on The Robot Report.