DREAMBot will develop a proof-of-concept edge-cloud application based on the MIDIH reference architecture (RA) to enhance the performance of the Omnit AMR; where both levels use machine learning approaches of a different complexity. The use case is based on standard cart handling procedures where an important task element is for the AMR to identify a set of carts as it approaches the cart parking area in a factory or a warehouse. The AMR operate in a workspace shared with workers and capable of docking with the carts regardless of their arrangements on the shopfloor.
Figure 1 Scenario for cart pose estimation using edge-cloud learning
Figure 2 Tractonomy Robotics Omnit V1 AMR
In Dreambot, we process data from onboard vision sensors, in near-real-time to generate a probability map of carts in the viewport of the robot as it approaches the parking area. This will be based on lightweight machine learning classifiers implemented and running on the onboard PC, forming the Data-in-Motion or Edge layer. The AMR uses the information to plan the approach to the carts. Once in range and while in motion, the AMR pushes 3D point cloud data to the Data-at-Rest level (cloud) where parallel pose estimation algorithms verify and return pose estimates to the AMR for capturing the carts. All edge-cloud processing occurs during AMR motion. The DREAMBot architecture uses technologies selected to be compliant or compatible with the MIDIH reference architecture. We use ROS2 for end-to-end data streaming middleware, providing data security while guaranteeing Quality of Service (QoS) during edge-cloud communication, even in very unreliable wireless environments. We aim to perform a set of experiments to validate this architecture in a realistic demonstration environment including resilience of the edge-cloud application to network latencies and losses.
Figure 3 DreamBOT solution architecture
The dual-level recognition application, proposed in DREAMBot, is crucial to TRACTONOMY’s value proposition – delivery of an affordable, environment-aware and easy-to-use robot. Therefore, implementation of DREAMBot will have not only technical, but also business related outcomes – including reaching out to new potential customers and establishing a demonstration showroom in TRACTONOMY’s facilities. We are now in discussions with another major retailer in Belgium and global e-commerce brand in the US. While we have a core product that is very good, we strongly believe that the DreamBot outcomes played a major part in our ability to tell the Tractonomy story and for customers to believe in our story.
