Visual SLAM promotes the evolution of service robots

▍No one doubts that VSLAM will be one of the keys to driving further iterations.

Mobile "machine" or robot?

As the number of robots deployed continues to grow, the defects that were "ignored" in the early days are causing increasingly serious negative impacts. On the one hand, although laser SLAM has been continuously evolving based on Cartographer, there has been almost no significant progress in recent years. Restricted by the fact that lasers can only build simple geometric environment models, they cannot "see" the real environment and lack the ability to interpret environmental content. There are natural bottlenecks in achieving interaction, intelligent obstacle avoidance, and intelligent operation, which greatly limits the intelligent upgrade of robots. As end users pay more and more attention to intelligence, coupled with the emergence of large models, upgrading new navigation technologies has become inevitable.

On the other hand, in terms of deployment, traditional marker positioning and laser positioning technology solutions not only have a long deployment cycle, high cost, and complex operation, but also require offline team support. In addition, they are also poorly adaptable to the environment. In the face of large spaces and high ceilings, environmental modifications are often required, further increasing the difficulty of deployment. At the same time, once the environmental layout changes, redeployment is required, which greatly increases the cost of use. In addition, the high cost of laser radar has indirectly led to the high price of robots, and the value of replacing human labor is not obvious.

An expensive robot not only requires long-term assistance and support from a team of engineers, but is also relatively unintelligent. Obviously, it is defined more as a mobile "machine" rather than a robot.

VSLAM is not just about low price

Under the multiple pressures of the market and the industry, the leaders have been trying new technical solutions. With the rise of VSLAM technology, people have foreseen the huge imagination behind it.

Humans can obtain external information through senses such as hearing, touch, and smell, but more than 80% of it comes from vision. Through vision, the human body can obtain information about the size, position, brightness, color, posture, and other information of external objects, thereby interacting with the surrounding environment. Robots obtain external information through various means: from the earliest collision feedback, to laser radar scanning, to a stereoscopic vision system based on the "human eye" composed of dual (multi) cameras, completing the upgrade from one-dimensional to two-dimensional, and then to three-dimensional.

By imitating human vision, the robot gains perception of space and possesses the basic capabilities and prerequisites of "human". This is the biggest difference between VSLAM and other technical solutions, and also its core meaning and value.

At the same time, the natural cost advantage of visual sensors enables visual solutions to achieve the same technical effects as laser solutions while reducing costs by at least 50%, which also means a faster return on investment.

According to calculations by industry insiders, in commercial cleaning robots, navigation BOM costs account for about 40%. If the robot price is calculated at a gross profit margin of 50%, the navigation module with commercial application accuracy has a BOM cost of about 6,000 yuan for the VSLAM solution, and the whole machine price is about 30,000 yuan. The BOM cost of the laser SLAM solution is about 12,000 yuan, and the whole machine price is about 60,000 yuan. Calculated based on a three-year service life, the payback time for the visual SLAM solution is about 10 months, with an average annual ROI of about 87%, and the payback time for the laser SLAM solution is about 1 year and 8 months, with an average annual ROI of about 27%.

Obviously, the VSLAM solution will have a greater chance of bringing about an inflection point where robot costs are lower than labor costs.

It can be seen that some leading manufacturers in the market have already turned their technology route to VSLAM. For example, Pudu has developed its own PUDUVSLAM+ solution, which is a new generation of VSLAM technology developed based on visual cameras and has been applied to its "Happy Delivery 2" robot. As one of the earliest companies in China to make layouts, INDEMIND is also at the forefront of the industry in visual research and development.

Since its establishment, INDEMIND has insisted on deepening its research on visual navigation, and has deep accumulation and experience in key technologies and product development such as robot navigation, obstacle avoidance, decision-making, and AI interaction. Based on existing industry needs, it has developed a commercial robot navigation solution "Commercial Robot AIKit" with VSLAM as the core, which has reached the industry-leading level in overall performance.

More mature, smarter, and a generation ahead experience

"Commercial Robot AI Kit" is based on INDEMIND's self-developed OSFusionAGI system. It adopts a multi-sensor fusion architecture with binocular stereo cameras as the core. It supports mainstream sensors of different categories on the market, and meets the development of core functions such as commercial robot navigation and positioning, intelligent obstacle avoidance, path planning, and decision-making interaction. It can be widely used in commercial service robot platforms such as commercial cleaning, hotel distribution, food delivery, and inspection.

In terms of applicability, the "Commercial Robot AIKit" is almost fully compatible with commercial scenarios, and has effective solutions for highly transparent materials such as glass. It has been deployed in thousands of different commercial scenarios around the world, without the need for scene classification. The built-in active fill light strategy can work without difference in special lighting environments such as direct strong light, no light source, and dimness, meeting all-weather operation requirements.

At the same time, INDEMIND has specially developed an intelligent decision-making engine to effectively improve the robot's decision-making ability. Based on the powerful perception and cognitive ability brought by stereoscopic vision technology, it can realize multiple key functions such as intelligent obstacle avoidance, active safety, intelligent operation, intelligent interaction, and intelligent elevator control. Taking intelligent interaction as an example, based on the three-dimensional map and AI recognition function containing three-dimensional spatial information and object semantic information, combined with the intelligent decision-making engine, voice control can be used to summon the robot, let the robot follow, specify the target area for operation, etc., greatly improving the robot's intelligent experience.

In terms of cost, the "Commercial Robot AI Kit" is based on a fusion navigation solution with visual sensors as the core and the advantages of the INDEMINDOS Fusion system. The robot development cost can be reduced to less than 2,000 yuan, and the cost of the complete chassis including navigation and batteries can be reduced to less than 5,000 yuan, which is a 60-80% decrease compared to the lidar solution, significantly reducing the robot development cost and cycle.

In addition, products equipped with "Commercial Robot AIKit" can autonomously build maps in real time, so new machines and new scenarios do not need to be pre-deployed and can be used out of the box. After the scene is changed, the robot can also update the map autonomously without relying on offline support teams, and remote areas can rely entirely on sales.

At present, INDEMIND has reached cooperation with many giant customers at home and abroad. The medium-sized cleaning robot Cobi18, which cooperated with the traditional cleaning manufacturer E, has been deployed in batches in more than a dozen countries around the world, and has achieved zero failure in the European and American markets, receiving widespread positive feedback. In the future, INDEMIND expects orders to exceed 100,000 units in three years.

Review editor: Liu Qing