Analyze The Core Technology Of Sweeping Robots?

As the "eye" and brain of the sweeping robot, the navigation system gives the machine the ability to perceive and act. From random inertial navigation to global planning laser navigation, navigation technology has undergone many changes, and then laser navigation has become the mainstream of the market, but with the significant increase in users’ demand for intelligence, visual navigation based on computer vision technology is strong Appears, the game of "old players" and "new entrants" is staged again. Laser and vision, who is the "future"?

       Laser navigation technology

Laser navigation is born out of the early positioning methods based on distance measurement (such as ultrasound and infrared). Laser distance measurement emits a beam in a specific direction. After the light bounces, it is captured by the receiver, and the distance between itself and the object can be calculated by time. Laser navigation obtains environmental information through laser sensors, and measures the distance between the machine and obstacles. After algorithm processing, a two-dimensional map is constructed to realize positioning and navigation.

The specific working principle of laser navigation is to present a series of scattered, accurate angle and distance information points through the environmental information collected by lidar, which is called point cloud. Laser navigation calculates the relative movement distance and posture change of the machine by matching and comparing two point clouds at different times, thereby completing its own positioning. The relatively simple principle and the characteristics of the laser sensor make it have the advantages of high efficiency, high precision, and resistance to interference.

       Visual navigation technology

Visual navigation, as the name implies, is to collect environmental information through visual sensors and build maps based on feature points or markers to realize autonomous positioning and navigation.

Studies have shown that 75% of the environmental information obtained by humans comes from vision, while binocular vision imitates human vision in structure, using binocular parallax to achieve depth ranging. In terms of working principle, binocular vision can obtain massive and redundant texture information from the environment, and has a strong scene recognition ability, which provides a prerequisite for the robot to realize intelligent decision-making. The two-dimensional environment information collected by binoculars can be used to generate a three-dimensional map with depth information using stereo vision technology, which can calculate the distance, volume and attribute information of obstacles in the area, so as to achieve positioning, navigation, path planning, obstacle avoidance and other functions .

At the same time, combined with semantic recognition, binocular vision can better understand the house layout and spatial structure, and realize intelligent interaction, such as target tracking and execution of specific instructions.

Both have their pros and cons

From the foreseeable future, users have always maintained very high expectations for the improvement of intelligent demand, which also points out the direction for the next generation of sweeping robots. The robot must simulate the "mind map" of the human brain and learn to think independently. Intelligent decision-making is the key.

However, laser navigation is limited by sensor properties, point clouds cannot distinguish texture information, and they do not have the ability to recognize scenes. Therefore, they cannot provide effective support in intelligent decision-making and intelligent interaction, and their intelligent scalability is insufficient. At the same time, due to layout restrictions, low obstacles are prone to detect blind spots. In actual performance, problems such as obstacle avoidance failure, false touches, and slow response often occur.