Shulou(Shulou.com)11/24 Report--

With the continuous expansion of the attention and application scope of commercial service robots, the interaction between human and robot has become an issue that must be faced. For example, how do robots walk opposite people in the aisle, how to avoid pedestrians who suddenly cross, how to deal with people who "jump the queue" or "overtake" from behind, and how to deal with people who suddenly change direction. There are many similarities between these problems and our driving on the road.

In fact, as a low-speed self-driving scheme, commercial service robots must solve the complexity and subdivision of the scene. In human-to-human traffic behavior, there are established rules in most cases, but human-computer interaction in a closed or semi-closed environment brings many new problems.

As the Purdue robot enters more and more scenes, it begins to show its strong technical strength. The complex environment makes Purdue robot "smarter" and "more stable". Multi-product integrated solutions are leading the technological trend of the global service robot industry in multiple dimensions, representing the latest technological breakthrough direction in this field. this has also helped Pudu gain a leading position in the market.

Dynamic upgrade of ▍ perception algorithm

In terms of core technology, the biggest challenge of many human-computer interaction problems is the dynamic awareness algorithm. For example, when a commercial service robot is driving in a crowded and complex business environment, navigation and obstacle avoidance need to take into account the variability of the dynamic environment to ensure that the robot can quickly detect objects and avoid pedestrians, pets and children in an appropriate way. This is very important for man-machine safety. At the same time, part of the environment also requires that customers and restaurant staff can easily use and interact with the robot, which requires the robot to be able to better understand the environment.

There is no doubt that the difficulty of dealing with dynamic objects is much higher than that of static objects because of the complex interactive dynamic environment. Because static obstacle sensing can only rely on lidar or rgbd and other sensors to achieve. Through these sensors, the robot can detect the existence of obstacles, as long as know the relative position of the obstacles and the robot, can achieve simple path planning.

However, when the robot enters the environment with large human flow and faces complex dynamic obstacles, if it still deals with the static environment and does not calculate the speed and position of the obstacles in the future, it is very easy to collide.

In fact, this highly dynamic environment is difficult to pass by waiting, but requires the robot to first identify the semantic information of obstacles, such as judging the type of obstacles: chairs, pedestrians, or tables. Secondly, if we know that the obstacle is a dynamic obstacle, we can obtain the speed and motion direction of the dynamic obstacle through continuous multi-frame tracking to predict the trajectory of the dynamic obstacle for a period of time in the future. Finally, in order to avoid the collision between the robot and the dynamic obstacle, the robot can avoid, make way, stop and other operations in advance according to the future trajectory of the dynamic obstacle in path planning.

At present, Purdue Technology has shown two ways to implement dynamic sensing algorithms in its products. For example, the Pudu Qiaole delivery robot adopts the implementation method of RGBD DL, which mainly uses the deep learning algorithm, such as convolution neural network (CNN), to train the model, so that it can detect and classify different objects, and realize object recognition by detecting the position and boundary box of objects in the image. This kind of technical scheme requires high computing power, so the cost is relatively high.

The Pudu Bella robot mainly adopts the technical scheme of single-line radar + RGBD, which mainly uses the random forest method to select radar point cloud clustering similar to human legs, rgbd point cloud clustering, according to the confidence of radar point cloud clustering and rgbd point cloud clustering, to carry out post-fusion processing, extract pedestrians, and then use Kalman filter to track pedestrians in multiple frames to calculate the direction and speed of pedestrian movement. And then predict the movement trajectory of pedestrians for a period of time in the future, this way has lower hardware restrictions and higher popularity, and it is also a more mainstream technical scheme in the market after the launch of Pudu.

As the earliest users and promoters of these two kinds of algorithms, the dynamic perception algorithm of Purdue robot still has significant advantages. For example, the Purdue robot is ahead of its peers in terms of the accuracy of predicting the trajectory of obstacles and the robustness of subsequent decisions such as avoiding actions based on prediction.

The value of dynamic sensing algorithm to the terminal is self-evident. It not only improves the security of scene movement, but also reduces the conflicts and troubles caused by collisions, and strengthens the robot's ability to deal with complex scenes. In the difficult scenes, Purdue robot has been based on the perception of the environment and site recognition, to achieve a more appropriate speed for the corresponding site: for example, it can slow down at the crossroads and so on.

After Purdue implemented two dynamic algorithms, the robot was tested and verified strictly in the environment, which involved 50 experiments on each environmental condition. In the assessment of these laboratory environments, the robot has demonstrated an impressive ability to detect up to eight pedestrians at the same time.

In addition, in the laboratory tests, the robot demonstrated its powerful ability by identifying objects within a maximum distance of 5 meters from its position, and when encountered dynamic obstacles moving at speeds ranging from 0.2 meters per second to 2 meters per second, the robot has shown an amazing 100% success rate in avoiding collisions. This excellent performance highlights the effectiveness of the dynamic sensing function integrated into these robots.

▍ ioT technology enables larger scenarios

Not only is the dynamic sensing algorithm with high reliability and robustness, with the continuous expansion of the robot category, the robot covers a larger area, the robot may involve more cross-space floors. Then, seamless integration with existing environments such as existing restaurants and distribution systems is essential for a smooth user experience.

This is due to the lack of clear rules to manage commercial service robots running in public space in many scenarios, so how to enable robots to better use automatic doors, ladder control, or remote paging to solve the unpredictability of the environment, ensure that robots can effectively communicate their intentions and deal with complex social situations. It is an expandable direction that the robot can further improve the dimension of the environment and scene in addition to the dynamic sensing algorithm.

Due to the relatively new category of service robots and the lack of professional integrators, many commercial service robot enterprises have independently achieved a lot of integration extension in technology. IoT technology is an important way for Pudu to further realize integrated product solutions, and these ioT subdivision technologies also provide Pudu with a more perfect overall solution.

For example, in the field of automatic door / access control, Purdue has developed two schemes: pure software docking and hardware docking. Technically, pure software docking Purdue mainly allows the robot to request the "opening and closing service" of the automatic door control system. The robot sends instructions to the access control system through MQTT, and the access control system controls the door to open. In terms of hardware technology, Purdue mainly adds a communication system with access control and robot, which enables the robot to communicate with Pudu self-developed access control module through Bluetooth, and the access control module controls the access control host to open and close the door.

In the field of ladder control, Purdue robot can automatically call the elevator, enter the elevator automatically and press the button on the target floor to arrive at the floor and leave automatically. The whole system completes the robot ladder ride process by simulating human behavior, and does not control or destroy all the internal systems that do not participate in the elevator. It is an independent and safe elevator couplet scheme, which is also very in line with the current domestic and foreign requirements for elevator transformation safety regulations.

Purdue robot has different ladder control technologies for different scenarios, such as the transformation technology of Pudu self-developed ladder control module for weak network or even non-net scenarios, which is compatible with ladder control technology and three-party cloud technology of many elevator brand manufacturers. can better complete the robot to ride the ladder. Not long ago, Purdue has cooperated with Otis, and Lightning Box successfully got through the ladder control technology in Japan and successfully integrated seamlessly with the elevator.

At present, many service scenarios have been willing to accept the use of business service robots to complete digital upgrades. As robots enter more and more scenarios, in order to expand the scope of commercial robot services, it is often necessary to make major changes to infrastructure and operations to meet the growing demand.

This not only puts forward the requirements of a certain technology for robot enterprises, but also needs to have the ability of the overall solution. It is not difficult to find that the technical problems involved in the service robot system are not small, the highly dynamic and complex human-computer interaction scene begins to put forward more requirements for the technical capability of the product, and the technology begins to show greater scene value.

But despite these challenges, researchers at Purdue Technology are actively working to overcome these difficulties, constantly raising the upper limit of the capacity of service robots and expanding the scope of application of service robots. The practice of Purdue Technology also shows that the leading enterprises with core technologies are continuously optimizing scene service capabilities through technology iterations and digging deep into technical details, and are beginning to be favored by more and more markets, which further promotes enterprises to pay more and more attention to core technologies, forming a "technology-market-technology" closed loop.

To a certain extent, this method also shows the logical relationship between the market and technology of China's emerging strategic industries such as robots, and shows a new technological model for Chinese enterprises to go to the world.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.