www.vicorpower.com POWER INNOVATION 15 www.power-mag.com Issue 2 2023 Power Electronics Europe the backside and two more along the skirt. In all, 21 lasers map the surrounding area every 25 milliseconds. That data is used to create a 3D map of the area around the robot out to a 100-meter radius, which enables the ASR to “see” its environment. Secondly, sonar sensors are located around the robot to provide proximity sensing that allows the robot to tell when something is physically close. This allows the ASR to take evasive action or communicate with an approaching person. GPS is included as a tertiary input for internal navigation and helps track the machine in the event that someone were to attempt to move or steal the robot. Odometry sensors calculate wheel rotation to indicate if the robot is moving or tracking left or right. And finally, an inertial measurement unit, or IMU, provides six-degrees-of-freedom spatial awareness to determine if the robot is upright or tilted, which could signal it has become stuck or immobilized. Together, the five sensor modalities paint an incredibly accurate picture of the robot’s geofenced environment and allow the ASR to successfully navigate complex environments while avoiding people, animals and objects. AI and edge computing process data at lightning speeds The ASR architecture is built so that the data is constantly streamed and recorded. That creates about 90 terabytes of data per robot per year, which no human could process. Data is stored for up to 30 days, which is the typical security industry standard for data retention. Data is stored both on the robot and in the cloud, depending on the situation. The difference between the two sets of data is associated with the threat level. If a threat is detected, the data is simultaneously sent to the cloud and triggers an alert. If the robot is just recording video as it patrols the park, then immediate access to the cloud is not necessary and the data is stored on a local hard drive. Artificial intelligence is used on the navigation stack and to analyze different parts of the video stream. For example, AI is used for people detection and to enable facial recognition systems to determine the similarity ratio of a detected face versus one in the user-generated database. That could be used, for example, to look for a lost child at a park or festival grounds. AI is also used for license plate recognition, especially because the 50 U.S. states have so many different kinds of plates, personalized and otherwise. Power efficiency is paramount to ASRs The intense level of computing, communications, and sensing places a tremendous burden on the ASRs’ power delivery networks (PDN). The PDN must be compact and have high efficiency. Because the ASRs have no airflow or venting, Knightscope went hunting for a pure conduction-cooled solution that could use the aluminum skin as a heat sink. The company adopted a Vicor DC-DC converter module (DCM3623) because its unique ChiP™ (Converter housed in Package) design was thermally adept and very small. The high DCM™ power density also helped with routing the wiring and cable assembly and increased battery efficiency, performance and runtime. As Stephens explained, “Unlike an electric car, the goal is not to maximize range. It’s more about maximizing robot run-time and minimizing the charge time, because it operates 24/7/365.” On the electrical side, the robot required isolation from all of the different power rails. Because there are so many sensors with different EMI signatures, the Vicor DCM helped minimize EMI and noise interference. Knightscope has a long roadmap of features and capabilities that can be added to the robots. In this case, Vicor DCMs provide a scalable platform and uniform height that obviates the need to change the heat sinking or mechanical components. “The more we’re able to reduce the burden on the battery, the longer run-time we will get,” Stephens said. “So, power’s always, always going to be a Knightscope ASR can patrol indoors and out with incredible dexterity because of the many highly-refined sensors. Together, the five sensor modalities paint an incredibly accurate picture of the robot’s geofenced environment and allow the ASR to successfully navigate complex environments while avoiding people, animals and objects.
RkJQdWJsaXNoZXIy MjQ0NzM=