Boxie Autonomous Robot: 33min Deep Dive | Marvelmind

0:01 Hello colleagues. Today we are giving a pretty deep introduction of our new robot, which you called Marvelmind Boxie. What is this robot for? First of all, it is for autonomous delivery, for research, for education, and overall development. So let's jump to the basics. The robot is about 31 centimeters by 31 centimeters. It has a height of, I don't know, 15... I will ask my colleague to bring a meter and we will show more details. The weight is 4.5 kilos, and it has the following kinematics. So it has two driving wheels. It has

0:58 two caster wheels, and the chassis is made of metal, so it's very, very rigid because its payload capacity is up to 10 kilos. What can it do? First of all, what is the robot? The robot must be capable of doing three major things. A robot must be capable to sense the environment. This is why the robot has plenty of different sensors, and sensor fusion in general is the key for any autonomous robot. Second, the robot must be able to process the information, to record it and recall it from memory from the previous past. And the robot must be able to actuate, to affect the environment around. In our case it's delivery, first of all, but not only.

1:56 It has plenty of other interfaces that we will be talking about, because this robot is developed to be expanded. So effectively it's a tractor. It's designed to have additional equipment installed on it. So what you see here is what we call the standard configuration. So it already has the base, it already has the Intel RealSense lidar and 3D cameras, and it has standard batteries. The base configuration, if you go to our website and check it, the base configuration would be the same but without Intel RealSense. And the advanced configuration would contain everything the same but additional batteries for more capacity inside, and quite a few other things like

2:56 stationary beacons included, additional warranty, and additional training when you need it and if you need it. So let's jump a bit back. What are those sensors? Many, many, many, many. Because the environment in which the robot operates is challenging, and that's exactly what differentiates the real robots from the robots which are used, let's say, for laboratories or for demo cases. Because the real environment in the warehouses, assembly plants, in their, let's say, manufacturing facilities, is very challenging. You can have noise, you can have lights, you have no lights, and the robot must be able to operate in all this environment using one, two, three, or may be more sensors using sensor fusion. So what are those sensors?

3:55 Of course, our own precise and new navigation system. So as you can see, there are two omni microphones. Each of the microphones, each of them, omni microphone has four microphones. So this is why they are covering 360 degrees in each microphone, and there are two. And the base between them allows you to measure not only very precise location but also the direction, because you calculate the position of each of this dot with centimeter-level precision, and then knowing the base you can know the direction where the robot is facing with a couple of degrees accuracy. And of course this works always together with the inertial measurement unit. So this robot contains several IMUs. Inside the robot there are two Super-Beacons effectively embedded.

4:53 And you can hear, you can see those microphones from the Super-Beacon, and you can see their transducers, which are used in this case as sonars, but they are part of the Super-Beacons which are inside the robot. So I am using the Super-Beacons and it's available, so you can use them. And as usually, I'm using in this case gyros. Gyro is drifting. In order to eliminate the drift, the paired beacons configuration cancels this. So it meant that you have at the same time super precise and super accurate IMU, and at the same time without no drift, because the drift is eliminated using the paired beacons configuration. So that's not different from our typical paired beacons configuration which we are offering.

5:50 It's a paired beacons. Or it's not different from our larger brother of this small girl. Our robot for industrial applications, the 100, it's implementation of precise indoor navigation and positioning, plus location and direction using the paired beacons. Of course, since this is a robot and we are all the time talking about sensor fusion, it is using not only this for positioning, but also odometers. So there are two wheels. Each of these wheels has embedded odometers, and those odometric data is available to you also through API, and of course is used by the robot for autonomous driving. So it sends a fusion between the positioning,

6:47 the IMU, and odometer. But we know that in many cases a customer prefers options—not necessarily a must, but an option—to use optical navigation. Yes, for this purpose we are offering Intel RealSense 3D. And lidars, lidars are obviously used for building the environment and knowing the obstacles, and 3D for localization and positioning. And together this is a perfect solution. But there's a big, big, big but between the optical-based solutions and indoor GPS solutions. With indoor GPS solutions, if you travel for example 50 meters, you can return to the same point with two-centimeter accuracy. It doesn't matter how much you travel; you can always return with two-centimeter accuracy. So that's

7:45 kind of absolute positioning. With optical tracking it's not absolute positioning. It's always relative to tracking. RealSense works very well and meets the requirements, and they give about the closed loop one percent accuracy. Should mean that if you travel for example 100 meters and return back, that officially you can be up to one meter away from the point where you started, which can be pretty much for the robot of this size and in general. So it meant that optical is very good for, let's say, traveling and overall traveling without additional beacons. Because for example our indoor GPS does require the network of stationary beacons around, but at the same time is not without limitation. So this is why we are offering you

8:44 a choice: indoor GPS only, indoor GPS with location plus direction plus odometry plus RealSense as optical to complement the sensor fusion. That's about the position of the robot, which is the key, key requirement for autonomous driving, autonomous navigation, indoor positioning. But you also have the obstacles around you. You need to overcome those obstacles, detect those obstacles, and drive around them. So we have several sensors handling this. Now, as mentioned already, RealSense does the job as well, but sometimes it's just too complex to implement. You really need a lot of calculations and a lot of software in order to overcome those. We have additionally 1D lidars which are used for obstacle detection. They

9:43 Can detect up to four meters in dark environment. In the typical environment, very bright light environment, it's up to one meter. We recommend at least that, of course. That is also available to you through the API. Everything I'm talking about is available for the self-driving, but also for you to develop your system on top of the robot through the APIs. If something is missing, especially in the beginning, let us know and we would enable this, because we are using this and we are happy to enable this API to you. So it means that you can build, for example, take over the control of the wheels. Yes, the robot drives using our own self-driving system and software, but you can do this by yourself. So you can use the robot as a ready-to-use chassis with the hardware, with the drivers, with the software, and many, many other elements like sensors.

10:42 But you can build your own self-driving mechanism and not use ours through the APIs. So I will probably repeat this, but remember: so lidar. There are several of them—two in front, three on the corners, let's say two on the corners and one on the side, and a couple on the back side. But those are optical. Sometimes you may face objects which are providing the obstacle but not visible to the optical lidar. For example, too reflective or too translucent, or just like a mesh. So you cannot detect a mesh using lidar, or sometimes you cannot. So that would be pretty difficult.

11:41 Help overall. As long as this same logic is applied for positioning, sensor fusion. But the same logic is applied for obstacle detection. So you have, once again, RealSense 3D lidar. We have our own 1D lidar, and we have also sonars. So all these together, combined, helps a robot overcome and detect objects in a real environment. Of course, additionally, there are some other sensors which you probably cannot treat as sensors, but they are. For example, you are measuring the current consumption, and the robot is driving like regular driving, and you sense the current which is consumed by the motors, and then suddenly, for example, you miss everything—you miss by sonars, you miss—

12:39 By lidar, 3D lidar, by everything. Objects like this—very, very difficult to sense because it's sharp, it's thin, it doesn't reflect back. And for example, you didn't notice, so the robot didn't notice, and the robot simply starts pushing. And by sensing the current, you can detect there's some change in the driving. So current sense, from this perspective, is also an element of obstacle detection and, as a result, avoidance. And of course, by definition, you use current sense in order to estimate your battery lifetime, your communication with the customers and the users, etc. All this is available. Sensor fusion, sensor fusion, sensor fusion—as a mantra, is key for any autonomous robots.

13:38 And we have plenty of different sensors inside. But sensors is one of the elements of autonomous robots, but not sufficient. So you must have processing power. So let's talk about the processors and the boards and the overall processing what you have inside. We have several layers and several boards. The key is a Raspberry Pi. So we have a very famous Raspberry Pi installed inside—this is a tiny, low-consuming Raspberry Pi Zero with wireless connectivity. So it automatically gives you a very strong Linux and Raspbian and ROS on a Raspberry Pi. And of course, you have interfaces available to you—they're the standard 40-pin interfaces, absolutely the same as—

14:37 You can see on the web. So they're available. Some of them are used internally, but the majority are used and available to you directly. So Raspberry Pi is the key, let's say, high-level computer and processing. We are using it, but the majority of power and memory is available to you for your external devices that you will be using along and on top of the robot. But that's not the only one, of course. We have our own boards inside. So these pins are for the odometer board, and this odometer board also has 40 pins, but those pins are, let's say, lower level. So, for example, you have equipment and you want to switch on or off some of your, for example, arm or your—

15:38 Camera or some gripper. So you can power it directly from here: 12 volts, 5 volts, and you can have the ground down. So it means that you can provide like 40 pins and then ground, switch on and off. So everything is from these pins and everything is, of course, API-available. Additionally, you have a display, which is basically an Android phone. So you have two high-level processors: you can run Raspberry with ROS, and you have Android. So it's up to you in terms of the end-user interface. Let me jump a bit faster forward. So there's HDMI—

16:36 And this HDMI is connected to the Raspberry, so it means that you can easily connect a mouse, keyboard, and HDMI and upload your software and your program directly to Raspberry and run it natively over there and debug everything you need. Even more, you can power, for example, a display or monitor from some of the sources or from external additional battery installed here, and you can even run with the monitor installed on the robot if, for example, your development requires this. But let's return back to the computer. So what computers are available to you? First of all, Raspberry as a high-level, then our odometry board, which is the main controlling board to all other boards, which gives you direct access to many elements—

17:36 Like power supply switches, like direct control of the data which is received from the Super-Beacons and many other lower-level elements than Raspberry. And on top of this, as a user interface, you have the Android phone, which provides many, many other things. But let's talk about this a bit later. About the chassis: as mentioned, the key application is delivery from point A to point B. So check our videos. It can carry up to 10 kilos absolutely without problem, and it's very rigid. It's made of metal—I think it's two millimeters, 1.5 millimeters anyway. So it's very, very rigid. It's designed to carry a huge capacity, but you know, recommended is up to 10 kilos. The battery inside is effectively limited—

18:36 By the maximum capacity which is allowed to be sent over the air—so it's 100 watt-hours. That allowed today to be shipped, so that's the maximum capacity we recommend to install. And that's the maximum capacity that has the advanced configuration. In the base configuration, it's smaller, and in the standard configuration, it's in the middle. What else? The chassis, as you see, has many holes. Those are not for ventilation; those are designed for your particular boards. What are those boards? The holes are already pre-designed. It's, of course, Arduino. Yes, it's pretty weird to install Arduino on such an advanced robot, but why not? So there are holes specifically designed for Arduino.

19:35 Whole specifically designed for additional Raspberry. We always recommend, if possible, install your own Raspberry. Take the power supply either from USB or from this source and run your program or whatever applications absolutely independently from our Raspberry. That simply simplifies their development. So Arduino, Raspberry, and Jetson. So you can install Jetson as well. You can install all of them or some of them, up to your choice. And of course, regular mounting holes just to install your additional equipment like arms or whatever boxes or whatever thing you will be carrying using the robot. If you need slightly different configuration, let us know because this is configurable. We can produce it on demand based on your particular needs and based

20:34 On your particular requirements. A bit more about the battery. Yes, so since the robot needs to carry pretty much and drive pretty long, effectively all this area is a battery. Battery designated to the batteries. The default battery is about 4,000 milliamp hours or amp hours. So it's about 50 watt hours overall capacity, and the maximum is 100. But the maximum inside, you can easily install layers of batteries, whatever, up to 10 kilos of batteries if you wish. So on top of this, then connect, connect internally. It's very, very simple. I will ask my colleague to bring the battery

21:34 And I will show how the battery. So it means that it would be very, very easy to connect additional batteries. If you need so, the battery is 12 volts, so it's pretty common voltage and easy to use and easy to handle. So, let's talk about interfaces. As I mentioned, interfaces is this. So before the interface, a couple words about the battery. So similar battery science held inside, and you can connect more and more by simply connecting like this. By connecting more, you can first of all install additional batteries inside like this until you fill it completely. I mean, that's inside, and there's a special holder that holds the batteries. Or if you need, you can install

22:35 Huge layer of batteries here. And then the capacity is, I don't know, 10th and 10th amp hours of 12 volts batteries. So about the interfaces. Interfaces is another very important element for any robot because the robot must sense the environment, must process the environment, and must output to the environment. The main output is the delivery, so it's basically driving the motors, but not only. So what are those interfaces that the robot has? Now let's again, since I already showed, it's USBs. USBs is the simplest, the easiest. These USBs are connected to Raspberry, but many other boards are connected to Raspberry as well. What are those boards? No, odometer, of course, of course, Intel

23:34 RealSense and Super-Beacon. So there is a pretty extensive USB network inside the robot and six external USB interfaces available here. And traditionally, one interface is available inside. If, for example, you decide to connect something which must be installed on the top and you don't want to use the back side of the robot, so you can put the USB over there. What else? HDMI. Very, very convenient to run your program on the internal Raspberry if you need. So then, of course, the power supply. This power supply is kind of final switch on and off when you store it. In normal conditions, we recommend never to switch off the robot because robot is robot, robot is always alive. And you can send the robot to sleep like

24:33 Any Marvelmind devices. You can send the robot to sleep and it will sleep, and then you can wake up. So you don't need to switch it off physically. But when you store it, when you send it, or when you ship it, then please use, of course, their external switch, which cuts off the electricity completely. Charging: regular charging, one amp. That's what we supply to you. And if you have a need for faster charging, then we supply an optional fast charger. This fast charger is around five to ten times faster than this one. So particularly if I have larger batteries inside, then the fast optional fast charger is a great option. We have a couple of buttons, but those are like mouse buttons. You can assign many different

25:30 Tasks to them. Currently, they're just switching between different modes, and this is kind of pause and sleep mode button. But those which are currently assigned through the API, you can assign nearly anything. So it's kind of functional buttons. When you don't want to control and you don't want to install anything else, you don't want to control through the Dashboard, so these are functional buttons which you can assign nearly any functions from the robot to these buttons. And of course, reset. There must be a reset always. Our robots are reliable, but the reset must be always. So what about their again? Let me remind that these are 12 volts batteries, so 12 volts chargers are provided

26:30 What are the interfaces? No, these 40 pins. You can see their 40 pins from Raspberry. I2C is available, UART, SPI, audio in stereo mode, and some general GPIOs are available there. In this, nearly the same but more. I already mentioned switching on and off the power for external devices. First of all, I2C to have a lower level control, for example, of sensors. So you can get the same data through the API either from their odometer board or from Raspberry. It totally depends on your application. It totally depends on the depth and the level you want to have towards the system. So I2C is available, UART is available, SPI is available. Powering: 5 volts, switch on/off; 12 volts, switch on/off

27:30 Ground, switch on/off. Something else: check the manual for more details. But that's a physical connectivity. Physical connectivity is, of course, great, but not everything. There's also wireless connectivity, and we have plenty of wireless connectivity inside the robot. Now first of all, Wi-Fi. So since we have Raspberry Pi Zero, it provides the Wi-Fi connectivity directly to the Raspberry Pi Zero, which is easy to use because, of course, Wi-Fi inside the phone as well. Sometimes it's convenient, and it's convenient for the applications which are running on Android. But for, let's say, more robotics applications, not interface-type applications, this Wi-Fi would be recommended. Bluetooth over there as well. On the phone, you have 2G, 3G, and 4G

28:29 And of course, you need to provide your own SIM card. So it's effectively owned. So regular phone with Android which you are using as a display and many other nice things. Like already mentioned, 2G, 3G, 4G connectivity. Additional Wi-Fi if needed. But also GPS. Yes, the robot is designed mostly for internal applications, but why not use GPS if it's there? So yes, it's available to you as well through Bluetooth. What else? Let's talk about sensors a bit more in detail. I already mentioned the sensors, but those sensors from RealSense, basically you can reuse them right away. Now first of all, you can use them here and connect your external devices through the USB, but you can use them as well on

29:29 Raspberry as a part of sensor fusion, but in this case, you need to operate pretty deeply to extract this data from their sensor fusion because we are using this by ourselves. If you kind of disconnect this data source, then you can do multiple things on your own on your external boards, but then you are responsible also for the sensor fusion in general. So your board must be doing the sensor fusion on top. What else is on the front panel? Let me run through them. The LEDs are used mostly for optical communication between the robots, the optical communication with the users, and every real light to help real sense in the dark environment.

30:30 Mentioned in the beginning, there is always a challenge with different types of sensors. When you have too noisy an environment, the ultrasound positioning may fail because it's just too noisy around, or if you run under the table and the stationary beacons are not visible, then you need to rely on odometer and real sense. In order to drive out both odometer and real sense, they accumulate their error, and then you cancel the error by using the absolute positioning of indoor GPS ultrasound-based system. So all together it works perfectly all by itself and may have some challenges depending on the environment. So the optical obviously has challenges when you don't have the light around and you need it, so these LEDs are providing the light. Sonars now, as long as I mentioned, typical sonar is nothing fancy. These are emitting, these are receiving.

31:29 So first of all, you can detect objects around, typical objects which are visible to ultrasound, and you can even slightly know the direction to the objects because you can emit only by this and listen by this and then know the direction to the object, or emit by this and by their time difference of arrival, know the direction of the object. So it gets additional data for obstacle detection, not only, you know, I'm hearing with this, so the object must be somewhere here. No, it's more precise than that. So summing up again, the robot is designed for delivering, for installing edition equipment, and then many, many, many different use cases. The, you know, really long tail of use cases.

32:28 Cases. It offers great flexibility in terms of connectivity through the high level, through the low level, through the powering. And oh, by the way, I didn't mention additional camera in the face. So the upward-facing camera which is used for, for example, for QR codes or for positioning. So it's of course connected with the Raspberry. So it's a standard camera provided. So it complements all other systems and it is a part of the sensor fusion for different uses. Rigid, capable for expansion, easy to operate, and flexible in general because we know that there are so many different applications that customers are using our robots for. It's flexible by design, but if you need something more, if you need something special, or if you like to add something, just let us know. We are flexible in that as well, and we are happy to build a particular customized configuration for this Boxie robot. So thank you very much.