AGV Perception in HD
We are rebuilding the AGV perception stack with
human-like spatial awareness for higher speed driving, uncompromising safety, and lifelong autonomy.
Our platform leverages the latest advances in autonomous vehicles, and extends these to industrial environments.
3D LiDARS and HD Mapping
Box Robotics is bringing HD maps and 3D LiDAR to the warehouse space. A strength of HD maps is they embed priors that improve AGV localization, planning, and object detection performance beyond what can be achieved by the sensors at run time. The high accuracy and measurement density of 3D LiDARs enables the creation of voxel layers in HD maps, and the precise localization of objects therein. Pricing for 3D LiDARs also continues to fall. For these reasons and more, 3D LiDAR and HD maps represent the future of next generation AGV perception.
Deep Learning and Scene Understanding
Existing AGV safety scanners cannot discriminate between the leg of a person and the leg of a table. This limited spatial awareness has significant safety implications if AGVs are to ever leave the guidepath. At Box Robotics, we employ state-of-the-art deep learning approaches to object detection and tracking. Our proprietary deep learning models and training data are optimized for the industrial space. Using both 2D camera images and 3D point cloud data, AGVs can achieve unprecedented levels of scene understanding for safer vehicle operation, and replanning off the guidepath.
The open-source Robot Operating System version 2 (ROS2) is a robotics middleware framework designed to address industry concerns. Of particular interest to us is ROS2’s suitability to address functional safety standards, its determinism, and its built in security. Much of the robustness of ROS2 can be attributed to being built on top of the Data Distribution Service (DDS) used in critical infrastructure systems to include defense, finance, transportation, and energy. In addition to its DDS foundation, ROS2 introduces its own unique concepts that adds to its reliability and suitability for inclusion in commercial products. Our software is built on top of ROS2 yet can easily integrate with non-ROS systems.
WHY HD MAPS?
At Box Robotics, we are rebuilding the AGV perception stack with HD maps. Our vision is AGVs with human-like spatial awareness for higher speed driving, uncompromising safety, and lifelong autonomy.
Higher Speed Driving
AGVs improve warehouse throughput by operating around the clock, and without distraction. However, AGV speeds are constrained to 2 m/s. By comparison, human operated industrial trucks are driven at speeds in excess of 5 m/s and with much higher agility. The speed governor placed on AGVs is not mandated by a regulatory agency, but is due to the robot’s limited perception capabilities which gives rise to safety concerns. However, if AGV speeds could be safely increased, vehicle throughput would also increase. As a result, the vehicle fleet size required for a given facility would be smaller. A 50% increase in vehicle speed could decrease facility installation costs up to 21%, while doubling the vehicle speed could reduce costs up to 35%. With typical facility installations costing several million dollars, these represent compelling savings for vehicle OEMs.
Based on 30 seconds for a pick or a drop, and 400 meters of travel for a pick/drop cycle.
AGVs nearly universally adopt safety laser scanners to meet the object detection requirements of ANSI/ITSDF B56.5. They are very effective in this role. However, since they are based upon 2D LiDAR technology that scans in a single plane, safety scanners are incapable of detecting overhangs, cantilevered obstacles, negative obstacles, or dropoffs. This lack of spatial awareness leads to:
Accidents resulting in both damage to vehicle and product
Reduced speeds capped to 2 m/s, limiting vehicle and facility throughput
Reduced availability due to a lack of confidence to replan around blockages on the guidepath
Our HD mapping platform will eliminate each of these concerns, while uncompromising compliance with ANSI/ITSDF B56.5 and ANSI/RIA R15.08 industry standards.
Box Robotics HD map generated online from an Ouster OS1-16 LiDAR showing a forklift with raised pallet.
Output from a SICK microScan3 safety scanner. The pallet is invisible to the safety scanner since it is raised above the LiDAR scanning plane.
Tom & John co-founded Box Robotics in 2019. They have 36+ years of mobile robotics experience, including 15+ years of continuous collaboration across 3 companies. Most recently, they served as CEO and CTO of Love Park Robotics (LPR). LPR was a 3D computer vision software company with a focus on industrial applications. The software developed at LPR runs on thousands of mobile robots operating in production environments around the world today. Representative applications developed at LPR include a real-time pallet detection system for autonomous fork trucks (now offered world-wide by ifm), infrastructure-free localization software for AGVs, industrial robot depalletization software, and an open-source SDK for ifm 3D cameras that is distributed as part of the Robot Operating System (ROS).
Chief Executive Officer
Tom is an experienced technology entrepreneur, roboticist, and software architect. Prior to co-founding Box Robotics, Tom was CTO, Robotics Perception at ifm electronic, gmbh where he helped to establish the robotics business of a ~$1B global corporation. Tom joined ifm in 2018 as a result of the acquisition of his previous company, Love Park Robotics (LPR), that he founded in 2011. He founded and organizes PhillyROS, the local technical meetup for users of ROS. Tom has Master's degrees in Computer Science and Mathematics from Villanova University.
JOHN SPLETZER, PHD
Chief Technology Officer
John has been working with mobile robots for over 20 years. His experience includes automated guided vehicles, self-driving cars, construction vehicles, and large-scale localization and mapping systems to name but a few. John previously worked in academia as an Associate Professor of Computer Science at Lehigh University, in industry as CTO of Love Park Robotics, and in government as a test engineer for the U.S. Army. He received his PhD in CIS from the GRASP Laboratory at the University of Pennsylvania.
2025 Washington Ave
Philadelphia, PA 19146