Robots Learning to Recognize Faces and Objects


Numerous customer service applications – from manufacturing to order fulfillment – require recognition of faces and objects. Customer service robots need to recognize faces and associate them with customer data to provide personalized assistance. Tasks that are simple for any human child had traditionally been quite difficult for robots.

Cloud-Based Recognition Platforms

Today, cloud-based systems like Amazon Rekognition enable enterprise robots to interpret their surroundings more accurately. Rekognition uses the Amazon Kinesis Video Streams robot OS (ROS) extension to stream live video data from the robots of AWS RoboMaker subscribers. It then identifies and tags the objects in the video data and returns this information back to those robots through the Amazon Rekognition ROS extension. Amazon provides a pricing example of 50 ROS-based logistics robots in a fulfillment center that use object recognition. If each robot streams 60 minutes of video data per month for object recognition, at a rate of 7.50 MB of video data per minute, the total monthly charges would be $360.38.

C2RO Engage is a robot-agnostic cloud platform that makes autonomous navigation and facial recognition services available for robots – even those with limited computational hardware (e.g., a Raspberry Pi). Customers sign up for a monthly subscription that provides access to multiple services, including C-SLAM, that allow robots to navigate complex and unstructured environments safely. Engage uses parallel processing in the cloud to enable a near real-time turnaround. The AI can recognize customers to personalize messaging, send targeted alerts, track visitor activity, and integrate this data with existing customer relationship management (CRM) systems.

CloudMinds develops and operates an open end-to-end cloud robot platform as a service that is an evolving “cloud brain” capable of operating millions of cloud robots performing different tasks. Examples include natural language processing (NLP), computer vision, navigation, and vision-controlled manipulation. The robots and smart devices are connected over secure virtual backbone networks in 25 countries to the CloudMinds AI service. This AI service supports face and object recognition and navigation.

Application Examples

Humanoid robotics company UBTECH has released several social robots for interacting with humans, including Cruzr. This is a cloud-based intelligent humanoid robot that interacts with face and voice recognition and uses voice and humanlike body language to communicate. The Cruzr platform includes an open and customizable software development kit interface to control body movement, expression, speech, lighting, and facial recognition. The platform also offers Big Data analysis and management of data gathered from customer interactions.

Turing offers a security robot called Nimbo that sits atop a Segway Loomo mobile robot base and uses AI for recognition of objects and humans. Nimbo then identifies and alerts staff if it sees a person or vehicle in an area that is off-limits. It automatically categorizes and stores video footage in the cloud for easy access, evidence collection, and filtering. Turing also offers video analytics that work with any streaming video camera to perform real-time automated detection of people, objects such as vehicles, and activity and then provides instant alerts. The company can integrate a security drone that works with its security robots to gain a bird’s-eye view and even pursue suspects that it identifies from the air.

RightHand Robotics provides work cells with a robot arm and its RightPick gripper to work through application programming interfaces (APIs) with warehouse management systems that tell robots what to grab. The RightPick system brain uses 3D vision and cloud-based machine learning to identify individual items. The modular grippers are also designed to work with existing robotic arms. This system takes in visual and with tactile information, then stores that data in the cloud to share with other robots on the warehouse floor and learn from experience. Such “fleet learning” is an advantage of cloud robotics. Every time a RightPick robot picks something up and puts it down, it collects data and sends it to the cloud. This helps the robots to keep learning, get better and better at identifying and picking objects, and adapt to new packaging and new items. The company uses a hybrid financial model where customers purchase the work cell with the robot and gripper, then subscribe to a robots as a service (RaaS) solution that provides the cloud platform, updates, and maintenance.

Sanbot offers cloud-enabled humanoid robots and software systems that enable remote education through an open API that can be programmed to help educate students of all academic levels. Its cloud-based facial recognition also helps track and record student attendance records and provide check-in and check-out records through the cloud to parents via the Sanbot mobile Q-Link app.

Privacy Concerns

Every technology has drawbacks such as data privacy, a key concern for cloud applications and platforms. Voice, face, and image recognition and location-based tracking collect user data for the purposes of training and control. Increased concerns about the security and misuse of private information make it clear that companies can no longer take privacy for granted.

From a regulatory perspective, regulations like the European General Data Protection Regulation (GDPR) are driving technology companies that offer AI-based services toward more responsible and transparent policies around data usage. Data privacy is now a key value proposition for companies in cloud services, and by extension, the cloud AI space. Customers will expect and demand more transparency and protection of company and consumer data by cloud AI providers as the industry matures and companies compete on issues like privacy.

Comments are closed.