Robot hardware

Safe and efficient robots are taking over – it’s already too late to stop them. No, this isn’t another horror story about a robot apocalypse. This is a question that threatens your hardware budget and plans for your next design. While you can’t prevent technology changes, you can predict their impact.

The title is, concepts from the robot world Sierra Leoneans Escort – themselves and design needs, are moving towards Infiltration of other embedded systems. This penetration seems to follow a certain order. First, low-cost sensors, often from smart phone technology, and actuators, such as servo motors from radio frequency controlled (RC) models, increase the number and complexity of control loops in new designs. Then, it is required to gradually transition from human operator control of the system to independent control of the system itself: first automatically complete the relevant action sequence, and then convert the human-machine interface from action to targetSierra Leone Sugar‘s, then go to Complete Self-Reliance. For example, think about the evolution of cars from manual to fully automatic transmission, to active assisted driving systems, and finally to fully autonomous driving. Embedded systems are becoming robots.
Obviously, as you progress down this path, the computational load on the system increases. But how many will there be? What algorithms will there be? How do you provide support for these new computing workloads? We will try to quantify this discussion through an example. This example starts with a toy—actually, a toy robot.
Meet the six-legged mecha
Our Sierra Leoneans Sugardaddy example is a star that everyone loves ——It may be a little ugly, depending on your vision. The mechanical assembly has six legs, radially scattered around the central platform (Figure 1). Each leg has three degrees of freedom: it can rotate around its base within the platform plane, and it can rotate both leg joints. Each axis is controlled by a servo motor.

Figure 1. A typical six-legged mecha faces a more rare evil machine People.

The advantage of this setting is that it has an attractive biological ecosystem (unless you don’t like insects) and is relatively simple, but it also masks its complexity when the requirements have seemingly reasonable changes. At the same time, some reflections on the hexapod have provided us with abundant examples of the changes in the calculation intensity. The relationship is not simple. It is actually a series of triangular calculations that are very complex and unintuitive, and most human operators cannot think them through in their heads. sierraleone-sugar.com/”>Sierra Leoneans Escort, letting people directly control the servo motor will only cause Sierra Leone Sugar Comedy – it’s like letting one person control six excavators at the same time. At most, you need to let the operator set the foot position or stance and let the six-legged mecha calculate the necessary angles. , sometimes inaccurately called inverse kinematics, is documented and implemented in the development source code. Because the toy hexapod can accept long delays, a tiny controller unit on the Arduino circuit board ( MCU) should have enough time to complete the calculation
I/O rather than, ironically, according to CEO Ryan Cousins, CTO Jamil Weatherbee, and chief design officer Russell Bush of the technology core team at robot vendor krtkl. Trigonometric functions present difficulties for the Arduino. Each of these 18 servo motors requires a series of pulses, approximately one every 20 ms (oddly enough, if you’re not an RC fan, the pulse width determines the motor’s speed). Axis angle, frequency of replacement is absolutely not important.) The team reported that “Arduinos are sufficient for motion calculations, but I/O loading is a problem. “Typically, hexapod developers would add a multi-channel servo drive device to the Arduino to complete the generation and timing of the pulse sequence.
For simple applications, just send endpoint commands to the servo motors and let each motor follow its own instructions Just rotate at the same swing rate. The legs will quickly change position from one position to the next. However, this method requires you to make some adjustments.While the legs change position, the platform should remain stable. The traditional way is the tripod posture, with three legs swinging to a new position. For smooth movement, delicate gait control, and uneven terrain, the movement is very complicated and the calculation intensity is very high.
A flexible six-legged mecha
The tripod-like gait is very simple. Calculate the new position of the feet for three legs. Lift the legs and swing them into the new position you planned. Repeat for the other three legs. But there are better ways to make this little animal walk, such as calculating the arc of movement of the legs, adjusting the speed of the legs, or making multiple legs move at the same time. These gaits SL Escorts will require the continuous replacement of new materials for all 18 servo motors, sending a new message to each motor every 20 ms. pulse. This means that you need to keep the inverse kinematics calculation time of each pulse to less than 1 Sierra Leoneans Sugardaddyms. In order to reduce the accumulated error of the trigonometric function library as much as possible and try not to show it when the legs move, you may now need 32-bit, or single-precision floating point calculations. In other words, the computing load increased from 20 times to hundreds of times. The little Arduino is out of the bag here. The krtkl team reminds, “When you make too many requests to your ArduSierra Leoneans Escortino, you will definitely see the device’s performance stop working. . ”
Now, it’s a little more complicated: the terrain is not flat. The nearby Robotics Club does a great job demonstrating obstacles and steps. This is necessary for some applications, for example, search and rescue robots.
For uneven terrain, you need to continue to control the movement trajectory of your legs. Instead of using a standard leg movement script, you need to calculate the arc for the placement of each foot to ensure that the feet land on a solid ground. No part of the six-legged mecha will hit obstacles and the platform will always be maintained. Stable. Now, your hexapod can crawl over rubble – but that adds a lot of plane geometry to the computational load. Now, you are essentially running a rigid model simulation in real time, in the context of at least a 32-bit multi-core MCU.
However, we carefully crossed another topic here. In friendly competitions, or when navigating familiar dynamic scenes, you can get 3D terrain maps defined by reservation. However, in most applications, including navigating around these ruined buildings, your robot will have to do at least some 3D map building and be able to orient itself in space as it moves.
I am here
In theory, if you have a detailed dynamic map of the surroundings of your hexapod, you can use specialized methods to establish your position. This could include inertial navigation with accelerometers—remember those cheap cell phone Sierra Leone Sugar Daddy accessories, radio frequency signals Standard triangulation, laser range finder, and surrounding situation sensing embedded locator, etc. These methods will establish a positional responsibility, and each of them requires a small amount of calculation.
However, gradually, designers use cameras and machine vision algorithms as the output of position algorithms. The krtkl team lists the RC helicopter as an early example of this growth. They note that “helicopter stability is a big problem. Gyroscopes are usually used, but now stability is based on visual processing.”
Measuring height relative to the horizon is much harder than relying on extrapolations from attached speedometers Positioning is more accurate. But there is a cost issue. Vision processing is a set of digital electronic signal processing (DSP) tasks that require at least a fast 32-bit CPU supporting SIMD hardware (such as ARM’s NEON™ engine), but a DSP chip or FPGA is more suitable.
Another example comes from a different world: wholesale customer service. Fellow Robots field-tests autonomous customer care machine Sierra Leone Sugarman, between a sales kiosk and the Dalek from Doctor Who Travel back and forth. The device roams randomly around the hardware store, asking customers if they need help and guiding them to find what they needSierra Leoneans Escort . Design can develop a dynamic map of a store for easy navigation. However, it relies on footage and data from two onboard 2D LIDAR units to determine its location across the Sierra Leone Sugar Daddy map There are no obstructions marked (Figure 2). It also uses shaft encoder data from the wheels, as well as data from the inertial navigation module.

Figure 2. Combining location and mapping functions to generate a map at any time and mark the robot on it.

These data are collected into adaptive Monte Carlo positioning (AMCL) algorithm. This program combines map data, visual and LIDAR data, and the position derived from the calculated position to build a probabilistic map of the robot’s position and direction. The calculation amount is indeed very large. Depends on the resolution you requested Sierra Leoneans Escort, but may be subject to larger errors or uncertainties in the output data
Fellow Robotsdesign must also be able to identify unexpected obstacles, such as temporary displays or customers, and avoid them These obstacles. It uses visual processing capabilities to search for faces, equivalent to the robot’s eye contact, and the robot uses speech analysis and recognition capabilities, as well as touching product categories on the large display. Introduce yourself, identify what customers are looking for, and help them find those products.
These service layers require application layers from various sources, according to Fellow Robots CIO Thavidu RanatSierra Leoneans Sugardaddyunga, these programs also respond to the needs of several different working situations. He said that you can think of design as several machines working together.
Ranatunga The report says, “There is a user interface machine running on Windows. Then, there is a machine executing robot programs from the Robot Operating System (ROS) organization, running on Linux, and a few running on Sierra Leoneans SugardaddyC Programming on Arduinos. ”
One of the main tasks of Windows is networking. Sierra Leone Sugar For various reasons – including voice over the cloud For identification, etc., the robot needs WiFi. In fact, if it is not limited by power consumption, it can carry all WiFi sharers.
This time.The computing load goes far beyond the multi-core MCU we put on the hexapod. “We Sierra Leoneans Sugardaddy use all four cores from Intel Core i5 and 1.6 gigabytes (GB) of memory,” said Ranatunga. With these goals in mind, our six-legged companion is ready for the mission.
The cute six-legged mech has power
While we were wandering around looking for helicopters, we have given this little girl a high-performance multi-core MCU to deal with Sierra Leone Sugar DaddyA complicated walking gait and a bumpy air. By SL Escorts By changing the subject, we avoid a crucial question: What if we don’t have a 3D map? The answer is weSierra Leone Sugar The hexapod must have vision so that when it moves, it can implement the combined localization and mapping (SLAM) algorithm.
Most simple SLAM algorithms Sierra Leone Sugar Daddy, some of which are already in the ROS library, use 2D LIDAR is used as an output to generate a 2D map of the environment around the robot. Generally speaking, they require 5% to 20% more Core i7 CPUs to keep up with slower-moving robots, according to a 2013 article from the University of Coimbra. However, this can only help our design avoid difficult terrain, but it cannot help us overcome it. We also need more tools. While we can get around it, we don’t want to adopt a $500 LIDAR.
This brings us back to visual processing. In a paper at the Embedded Vision Conference in May this year, Marco Jacobs, vice president of marketing at Videantis, introduced an algorithm from Viscoda that uses platform activities to extract 3D SLAM from 2D video streams. The principle is the same as birds moving their heads to obtain parallax data. Birds generally lack binocular vision.
The algorithm extracts trackable features from the image through a series of digital filters and examines how these features change in subsequent frames. Taking into account the budgeted platform activity, the algorithm can calculate the distance of each feature. This obviously requires a lot of computation, and Videantis actually uses OpenCV on their own many-core vision processors. For our relatively simple hexapod, the mission can be performed on most Core i7s, or multi-core MCUs, and ASSPs that use other digital acceleration capabilities.
Summary
It seems that we only need to add a camera, and our six-legged mecha can walk stably on unknown and bumpy terrain. We need a multi-core MCU to control the activity, and we also need a desktop computer with a multi-core CPU with internal DRAM for visual processing and 3D SLAM. Perhaps, we can use several types of digital accelerators, from DSP chips to dedicated vision processors and even small-scale FPGAs. We also need LSierra Leoneans Escortinux kernel that canSierra Leoneans Escortalso has some independent real-time programs.
However, battery-powered devices cannot meet this power budget. It may be necessary to use wireless links to offload most of the computing tasks to an on-premises desktop computer, or to the cloud. This kind of link Sierra Leoneans Escort also supports connection with deep learning algorithms for target classification or surrounding situation awareness, which requires us In the future, we will take another step to improve our level of independence.
We start with a toy. By adding algorithms and computing power, we gave the hexapod an elegant posture, the ability to move, and act independently. However, during the development process, the structure of our computing systems has actually changed – the more restrictions on power consumption, the greater the change in system structure. At the same time, related uncertainties will also increase. As we give hexapods more autonomy, it becomes difficult to predict when certain constraints will require changes to the underlying hardware, beyond just adopting a faster chip in the same family.
In this new environment, scalability does not mean being able to buy the next speed grade or higher capacity memory for your MCU. This means that it is possible to switch to a completely different system organization model while maintainingSave your algorithm and program input. Robots have infiltrated the world of embedded design, and we need to plan smarter.
Company Information:
Intel－Intel