In recent years, we’ve got used to seeing robots that can walk, do parkour or even dance. However, it is important to remember that those robots are usually performing extremely well under certain constraints and with a little control from some human beings. The most difficult step for humanoid robots is for them to be able to understand autonomously the real environment in which they operate, and then take an appropriate decision about their next move. That’s why many researchers around the world are still working to improve navigation skills of machines to enable them to become more autonomous under any circumstances.
MOVING CAMERAS TO INCREASE EXPLORATION CAPABILITIES
In fact, also our colleagues from Rome laboratory, in the last year, have been collaborating with the Humanoid Sensing and Perception Centre of Istituto Italiano di Tecnologia within a project aiming at increasing the capabilities of robots to move in an unknown environment. “Our strategy actively combines RGB-D images, obtained with a depth sensing device associated with a RGB camera, with 2-D LiDAR data,” explains Francesco Puja, Research Technology Manager at Konica Minolta Digital Services R&D. “By active combination we mean that the robot can move its head to fully exploit the RGB-D camera. Therefore it can detect, and hence avoid, obstacles undiscovered by the 2D LiDAR, as such as overhanging obstacles or obstructions in blind spots. This approach can be used not only with robots, but with any moving camera platform to increase the awareness of the environment, detecting different situations and responding to different real needs.”
NAVIGATING IN REAL ENVIRONMENT
“We propose a strategy that integrates LiDAR and RGB-D information acquired respectively from the robot base and the robot head. The idea is to actively control the head to perform smart observation of the environment to inspect salient areas that are either outside the LiDAR field of view or that contain potential obstacles”, comments Lorenzo Natale who is Principal Investigator at IIT. “So, overall, our robot achieves a better understanding of the context around itself, and is capable to detect obstacles and navigate better within an unknown environment”. This approach has been validated in both simulated and real setting demonstrating the advantages of the proposed methodology, results have been recently presented at I-RIM 3D Expo 2020, a conference organized by I-RIM (Istituto di Robotica e Macchine Intelligenti).
Even if the demonstration activities have involved the use of the R1 robot, a humanoid robot developed by IIT, the approach developed is independent of the type of 2D LiDAR and RGB-D camera and from the navigation stack used, thus it can be applied to any robot with a limited field of view (FOV) and to different path planning algorithms. Indeed, the solution has been made compliant with the Robotic Operating System (ROS), a flexible framework for writing robot software.
FROM ROBOTS TO MANUFACTURING, LOGISTICS AND RETAILS
Within the next months, following up on these activities, our laboratories will keep developing solutions for automatic video analysis within the framework of a wider Konica Minolta project that involves several teams from research to business and sales units. In fact, since 2020, our R&D projects supported the integration of AI algorithms within the solutions based on Mobotix cameras to improve their performances in a specific context, as such as for digital manufacturing, logistics and retails.
For instance, Box Defect Detection is a system, soon to be available as part of Konica Minolta IT Services that locates all defects and anomalies of delivery boxes running on production line. The tool identifies defects’ size and severity and commands the separator on your production line to send the defective box to the washer or out of the line instantly. Suspicious Behaviour Detection is another application of Video Solution Services (VSS) recently developed for one of our customers that is managing printing services for exams, in the interest of the Czech Government. In this case, the solution can automatically detect suspicious behaviour, creating a security area around Konica Minolta’s production printers or in a restricted area.
Within the framework of VSS, we are developing other projects:
Vehicle occupancy detection is a solution developed for clients with high-security standards at facilities access control. It is a system that automatically detects and counts the number of people inside a car, denying access if more than one person is inside the vehicle and communicating the issue to the security guard.
The Occupational Health and Safety bundle is a set of video analytics solutions developed to ensure health and safety in workplaces such as manufacturing plants, warehouses, hospitals and more. For instance, the system can automatically detect when people are not wearing Personal Protective Equipment (PPE) as such as masks, glasses or helmets, and it reports the violations to the security personnel. They can recognise different kinds of activities or verifying people are keeping the appropriate distance. Other services include acoustic surveillance, temperature screening and intrusion detection.
For more information on Video Solution Services, request a pilot with our researchers.