From Single To Fusion, Who Is The Optimal Solution Of The Sweeping Robot

Under the wave of artificial intelligence, smart home products emerge in an endless stream, but throughout the development of the industry can really walk into the family of products on the count, and sweeping robot is one of the "Internet celebrity products".

According to data from Zhongyikang, the market scale of sweeping robots in China was only 840 million yuan in 2013, but by 2020, the market scale has reached 9.4 billion yuan. The rapid growth of the market is also forcing the performance of the sweeping robot to continuously improve the "lower limit", especially intelligent has become a high value-added feature of the sweeping robot.

Step by step, the sweeping robot is gradually "perfect"

After the cold wave in 2019, the sweeper market rebounded rapidly in 2020. In addition to the impact of the epidemic, the most critical reason is still the innovation and progress of sweeping robot technology. In particular, a variety of sweeping robots with sweeping integrated, automatic dust collection, automatic cleaning mop and other functions appear, greatly improve the cleaning ability, to solve the user has been criticized by the sweeping robot "not sweeping the floor" awkward situation.

And sweeping robot another big pain point, intelligence is also focused on the industry. Since the development of sweeping robot, the iteration of navigation technology has gradually changed from bold innovation to progressive innovation. Although the underlying technology is constantly enriched, it is difficult to significantly improve the performance of existing products in a short time. At present, in the intelligent upgrade more targeted to improve the product's intelligent obstacle avoidance ability, human-computer interaction and so on.

From singularity to fusion, who is the "optimal solution"?

In 2020, the star products in the sweeping robot market, Stone T7Pro and Covos Geopo T8, both adopted the way of Lidar fusion vision sensor in navigation technology, which not only broke the long-term competition situation of serious homogeneity, but also received wide praise from consumers in the consumer market. From a technical point of view, the reliability of Lidar has been widely verified, but its shortcomings are also increasingly prominent with the increasing market demand. Due to the limitation of sensor attributes, the environmental semantics cannot be identified, and the layout is limited, resulting in the actual obstacle avoidance effect of the product is not ideal, and there are often miscontact, collision and other phenomena. The addition of vision sensor greatly makes up for the defects of LiDAR, and significantly improves the obstacle avoidance ability and decision-making ability of the product. In addition, it can also realize remote cleaning through the visual sensor, and with the voice can also complete a series of interactive actions, more playable and interactive.

Pioneer products have been verified by the market, making this kind of products quickly rich, but in the process of the laser sensor layout restrictions caused by the edge of the problem gradually into the consumer's vision, due to the high limit, resulting in most of the bottom of the home difficult to clean or often stuck has become another pain point. Earlier this year, the X100 sweeping robot released by 360 and the ultra-thin sweeping robot released by Xiaomi not long ago are clearly targeted to address this need.

In order to solve the constantly upgrading market demand, technology fusion programs are becoming more diverse and mature, but the cost surge caused by it cannot be ignored. At present, most of the applications of such programs in the market are high-end products, while mid - and low-end products are still difficult to bear. Secondly, with the continuous addition of sensors, not only brings challenges to the cramped robot space, but also significantly increases the complexity of the system, resulting in more complicated calculation pressure, and constantly improve the platform computing power, will only lead to further increase the cost. High costs are clearly at odds with sinking market trends and are not universal. The role of visual technology in navigation, which is often the "supporting role" of convergence solutions, goes far beyond that.

Visual maturity, from "supporting role" to "leading role"

In fact, vision is not a new technology. It has also experienced years of development with sweeping robots. However, due to the difficulty of visual technology development, the early product application performance is not ideal. With the continuous maturity of visual technology, it shows great potential in navigation, obstacle avoidance, identification, interaction and other aspects as well as its easy integration degree let the industry see a new direction. As an early AI technology company with computer vision layout in China, INDEMIND specially developed a binocular vision navigation scheme for sweeping robots in 2020. Compared with the technology fusion scheme, this scheme only costs 1/3 of its cost. At the same time, the layout of the minimal restrictions, can greatly reduce the height of the fuselage.

In terms of function implementation, the scheme can provide map construction, path planning, object recognition, intelligent obstacle avoidance and decision interaction and other functions, to achieve navigation, obstacle avoidance and decision ALL IN ONE, significantly improve the integration and intelligence level of the system, reduce the research and development cycle.

In addition, positioning accuracy < 1% and attitude accuracy < 1° have been achieved in the navigation and positioning data, reaching the same level as the laser scheme. Relying on unique stereo vision technology, it can provide depth calculation with error less than 1% within the range of 0.05-1.5 m, realize three-dimensional map construction, can identify more than a dozen categories, hundreds of household items, object recognition accuracy < 2cm, and according to the identification information can make different obstacle avoidance distance avoidance action. At the same time, multiple interaction requirements such as target tracking, designated area operation and care can be achieved by configuring business logic to improve the robot's intelligent interaction ability.