How hardcore can a sweeping robot be? It has all the hard technologies such as autonomous driving, laser scanning, and NLP. Ecovacs: Only technological innovation can break the industry's internal competition

Mingmin Xiaoxiao sent from Aofei Temple
Quantum Bit Report | Public Account QbitAI

What? Your impression of sweeping robots is still that they spread feces all over the wall?

Nonono, the closest thing to autonomous driving right now may be your own robot vacuum cleaner.

In addition to scanning the room and building its own map, it can also understand the "topography" of your home better than you do:

You can also scan directly through the camera to get a 3D map with depth information, just like Apple iPad Pro:

It’s not even just a 3D map that scans shape and depth, but also a scene understanding map with its own labels:

Yes, this is indeed what the current generation of sweeping robots looks like. Not long ago, the latest sweeping robot released by Ecovacs, the Deebot X1 , has all of these abilities already realized.

How smart is it?

You only need to say "OK YIKO, go clean under the sofa" to let the robot vacuum cleaner accurately complete the owner's task:

△ Voice Assistant YIKO

This can't help but make people curious: from artificial intelligence to "real intelligence", from a semi-automatic "tool man" to a "housekeeper", what has the sweeping robot experienced?

Why is it so difficult to keep a sweeping robot from bumping around?

In fact, there are many issues to consider if a sweeping robot wants to complete a cleaning task.

What things need to be cleaned? What things need to be avoided? At what distance should they be avoided? How to clean the ground around the items after avoiding them? What to do if you encounter people or pets walking around...

What you see as a “simple sweep” actually requires the robot vacuum cleaner to make sweeping, mopping and obstacle avoidance decisions more than 100,000 times .

In order to better cope with complex cleaning situations, the " eyes " and " brain " of the sweeping robot have achieved super evolution.

Now, the sweeping robot can not only "see" items like slippers, socks, and wires, but can also identify them at a distance of about 2-3 cm, and then complete a cool avoidance.

Behind these small actions, a series of technologies that are also used in aerospace, autonomous driving and smartphones are used, including LDS laser radar, 3D structured light, dToF, AI algorithm, etc.

Take the “avoid slippers” operation for example, it is not easy for a sweeping robot to do it well.

Among them, just to identify objects, a combination of 3D structured light, dToF, and AI algorithm technologies is required.

3D structured light technology allows the sweeping robot to "see" objects and "understand" the environment, that is, to collect information such as the depth and position of objects through sensors.

By building a dot projector into the sensor, it emits thousands of tracking points to the object to be identified, and then uses an infrared camera to take a picture of it.

By using the data collected by the infrared camera and combining it with the built-in algorithm, the 3D information of the object can be quickly obtained.

The facial recognition we commonly use is based on this technology. If you have an iPhone, you can see the dot matrix projector and infrared lens required for 3D structured light from its small bangs .

At this point, the sweeping robot is only doing preliminary work.

It is not enough to simply obtain three-dimensional information about the roadblock. Today's sweeping robots must figure out what the roadblock is.

This requires AI algorithms to come into play, allowing the sweeping robot to "understand" objects and achieve classified cleaning.

Through target recognition algorithms, the sweeping robot can classify and judge common objects in the home environment.

For example, small debris and carpets are items that need to be cleaned, while slippers, socks, electrical wires, etc. are objects that need to be avoided.

Although these steps seem very simple, they actually require very high computing power from the robot.

Like the latest Deebot X1 released by Ecovacs, in order to be able to recognize more objects and make faster responses, it uses Horizon's dedicated AI chip at the intelligent driving level.

You should know that the computing power of this chip can reach 5TOPS . According to Horizon's official propaganda, its capabilities can reach the L2+ level of autonomous driving.

Therefore, with the sweeping robot equipped with autonomous driving technology, it will no longer "smear feces on the walls" or run around the house dragging socks.

After being able to see and recognize objects, the most critical step comes: avoiding them .

From the action of avoiding the slippers, it can be seen that the sweeping robot did not avoid the obstacles after hitting the slippers, but judged the obstacles in front when it was still a few centimeters away from the slippers. The whole process was very smooth.

This shows the magic of 3D structured light combined with dToF technology.

3D structured light can obtain the object's 3D information as well as its depth information, but the distance it can achieve is relatively limited; dToF can supplement the depth information at a long distance.

It transmits and receives multiple light signals to the object within a single frame measurement time, records the flight time of each signal, and makes a histogram. The flight time (ToF) with the highest frequency is used to calculate the distance to the target.

△ Image source: Wikipedia

This is why the sweeping robot will brake and readjust its route in time after reaching a safe distance from an obstacle.

Moreover, with the help of AI recognition, the sweeping robot can also make different avoidance strategies for different types of objects.

For example, for safe items such as slippers, socks, and chairs, it will get as close to the object as possible and will adjust its route only when the distance is about 3cm; for more dangerous electrical wires, the safe distance will be increased to about 5cm.

Sweep, but not completely? No longer exists

Do you think this solves the problem?

In addition to the obstacle avoidance technology that is shown off, a sweeping robot must have a full understanding of the home environment and clear route planning in order to complete a perfect cleaning task.

Snake - like movement vs. planned route

This requires looking at the LDS laser navigation and dToF technology that allows it to get rid of the "snake-skin movement" .

A rotating laser radar is hidden in the small disc on the top of the sweeping robot.

It can emit lasers while rotating at high speed. Based on the time it takes for the laser to reflect back after irradiating the target object, it can determine its own position and achieve positioning.

Like the random sweeping robots that had no navigation before, they often swept wherever they went without any planning at all.

The end result is: the floor was swept but not completely, and the owner had to sweep it again himself.

The dToF technology can further improve the accuracy of navigation.

Its biggest advantage is that it is almost not affected by light , which can ensure that the sweeping robot can still maintain super high accuracy when working in a dim environment.

In fact, today's sweeping robots will conduct a comprehensive scan of the home environment and build a 3D map before working .

At this time, we still have to rely on the powerful combination of LDS laser radar and dToF.

With the help of SLAM algorithm, the sweeping robot can start from any unknown point in the home and identify the home environment information in 360°.

It can also build an accurate 3D floor plan, identify complex floor plans, and plan cleaning routes.

In subsequent daily use, it will further refine the already constructed floor plan, re-correct any deviations, and ultimately allow the robot to thoroughly “figure out” the terrain of the home.

In addition, LDS laser navigation allows the robot to locate itself in real time and record it.

On the one hand, this allows the robot to clean according to the planned route, and on the other hand, it can continue to return to the place it has not just cleaned after charging, avoiding the missed cleaning and repeated cleaning that would occur in previous random cleaning robots.

It also breaks these "stereotypes"

BUT, as we all know, the inhuman "skills" of the sweeping robot are not just bumping around and running around.

When I was sweeping, the dust box was full, so I stopped sweeping immediately; I mopped the floor, carpet, and cement together to speed up the replacement; I accidentally forgot to set up the virtual wall, so I just went out and walked around the community a few times...

But now, these have become "traditional impressions".

In the case that personalized cleaning cannot be achieved without virtual walls, the latest sweeping robots can meet the cleaning needs of any location through AI semantic segmentation, 3D target positioning and recognition, combined with path planning technology.

In order to make the interaction more direct, the AI ​​intelligent voice interaction function is now added , so you can assign it to work by speaking. Especially for the elderly, the sweeping robot is much easier to use after it has voice control.

For example, if you don't need a sweeping robot to clean the whole house and want it to clean up the dog food under the sofa or the oil spilled in the kitchen, you only need to call the voice assistant named YIKO, "OK YIKO, go sweep the sofa", and the robot will understand the commands of "sofa" and "sweep" and arrive at the designated location accurately to do the work.

In response to the situation of "automatic strike", the latest sweeping robot base station has integrated functions such as automatic dust collection, automatic mop washing, sterilization and drying, truly achieving the effect of "completely freeing your hands" during the entire cleaning process;

There are also cases of indiscriminate mopping. The latest sweeping robots are equipped with AI automatic recognition, which can accurately identify the material of the floor and choose the most appropriate cleaning method. For example, they can mop on wooden floors and tiles, while sweeping on cement floors and carpets...

The latest Ecovacs Deebot X1 has basically solved all the complaints that players of the first generation of sweeping robots can think of.

Accumulated technology explosion

Looking back at the industry itself, sweeping robots seem to have just caught up with the era of technological explosion.

Just like the rise of smartphones back then, the technologies that can be installed are inexhaustible. Whoever can install more cutting-edge technologies at the same time will have a greater advantage in the market.

But did technology really explode without warning?

actually not.

As a type of technological product, the development of sweeping robots has gone through two decades and two iterations.

From the "novelty" of the 1.0 era, to the technological breakthroughs of the 2.0 era, to the true intelligence of the 3.0 era, the sweeping robots of each era have undergone very significant changes compared to the previous generation.

The 1.0 era refers to around 2001 to 2010. At that time, sweeping robots were not intelligent, and the built-in algorithms could only meet the most basic sweeping requirements. They could only be turned on and off through physical buttons, and had basically no other functions.

△ Like a sweeping robot from the 1.0 era

Although technologies such as inertial navigation and ultrasonic navigation have emerged, their shortcomings are also obvious. The hardware (gyroscope, accelerometer, etc.) that inertial navigation relies on is not accurate enough and still relies on collision algorithms; ultrasonic navigation is too susceptible to noise interference, and sweeping robots as a whole are in the stage of accumulating technology.

At this stage, there is indeed a group of consumers who are willing to take the first step. Their supporters are often a minority who believe that technology can change the world, and the general public may not necessarily buy into it.

The 2.0 era appeared around 2010, during which time AI algorithms based on deep learning began to emerge, and sweeping robots also began to achieve technological breakthroughs.

Around 2010, the cost of LiDAR technology dropped sharply, and the LDS-SLAM positioning and navigation technology for sweeping robots with LiDAR as the core began to appear, solving various problems that occurred in inertial navigation and ultrasonic navigation, and providing a mature solution for active navigation planning.

Around 2015, the industry began to see the emergence of the "sweep and mop in one " technology that combines sweeping and mopping, overturning people's impression that sweeping robots can only sweep the floor. Today, this function has become the wealth code with a market share of 92% ;

In early 2017, AI-based visual algorithms and interactive methods began to enter the robot vacuum cleaner industry, further improving the robot vacuum cleaner's object recognition capabilities and truly shortening the communication distance between machines and humans...

During this period, sweeping robots have a certain degree of practicality in terms of intelligence, functions and interaction, but most of them are still in a semi-automatic state. Although the number of buyers has increased, it still cannot make you who stays at home completely relaxed in terms of cleaning.

Now, with the launch of the Deebot X1 by Ecovacs on September 15, sweeping robots have also entered the 3.0 era , which is the ideal form of what people call "future robots".

Whether in terms of intelligence, interaction or function, it is becoming more mature. The robot's vision has both 3D depth and object information. The interactive capability and technology are deeply integrated, allowing the machine to have "thoughts" and complete the entire cleaning process fully automatically.

During this stage, sweeping robots have truly stepped out of the closed loop of " technology defines demand " and facilitated people's lives like smartphones.

In the gradual iteration and change of sweeping robots, Ecovacs has always been at the center of the wave.

As early as 2009, Ecovacs entered the field of sweeping robots as a domestic pioneer. At that time, the industry was still in the 1.0 stage, but Ecovacs had accumulated dozens of technology and design patents.

In 2013, Ecovacs became the first domestic company to apply LDS-SLAM to the field of sweeping robots, leading the industry into the 2.0 stage and embracing technological upgrades.

In 2015, it was the first to launch the sweeping and mopping technology, pushing sweeping robots into the era of sweeping and mopping integration; in 2017, it was the first to launch the 3D-drive three-dimensional cleaning system...

Ecovacs has many such "industry first" experiences: laying out the dual technology routes of VSLAM and LDS-SLAM ; launching AIVI, an AI system for obstacle avoidance; applying TrueDetect 3D structured light obstacle avoidance technology to sweeping robots...

From LDS-SLAM, sweeping and mopping to AI obstacle avoidance and interactive algorithms, Ecovacs' continuous breakthroughs in new technologies have endowed sweeping robots with more practicality and truly achieved a certain degree of automation, which is also the symbol of the sweeping robot 2.0 era.

After that, Ecovacs did not stop moving forward. In 2020, its investment in R&D alone reached 338 million yuan ; in the first half of 2021 alone, it had invested 202 million yuan in R&D .

As mentioned before, Ecovacs’ latest product, the Deebot X1, opens a new era for home service robots.

If the 2.0 robot vacuum cleaner can quickly plan and complete the cleaning of the whole house, then the Deebot X1, as a robot vacuum cleaner, is more like a smart butler:

The positioning, navigation and planning upgraded to 3D mapping are very different from the 2D effects of lidar mapping in the 2.0 era. Robots can now truly “see” the world as humans see it.

The use of AI voice interaction technology in sweeping robots has truly made the robots "understand human language" and perform tasks required by humans. This is the first time in the industry.

The cleaning function of the sweeping robot has also evolved from a simple "sweeping and mopping in one" to a robot that can automatically replenish water, wash mops, sterilize, dry, collect dust, etc.

That’s right, from the perspective of ordinary people like us, Deebot X1 is closer to what we consider to be a “ real lazy man’s artifact ”.

Unlike previous generations of products, the era of sweeping robots 3.0 has arrived.

Looking back at the history of sweeping robots, from the 1.0 era that was regarded as "useless", to the comprehensive technological breakthroughs in the 2.0 era, and finally to the "comprehensive intelligence" of the 3.0 era, it is also the result of the continuous accumulation of intelligence, functions and interactive technologies in the evolution.

Among them, Ecovacs is not just a player who happened to seize the node of the times, but a pioneer who dared to take the lead, dared to overcome difficulties, and launched technological innovation behind the scenes.

These changes cannot be achieved overnight; they require an evolution from quantitative change to qualitative change. Although there are doubts and ridicules in this process, those who persist in deepening their technological expertise and constantly pursuing product excellence will eventually become pioneers in a new era.

How convenient and comfortable our home cleaning will become in the future will depend on the hard work and exploration of these technology pioneers.