Resources

Building scalable robotics in a better way

Why general IPCs for AMRs are expensive and troublesome

It’s the era of AI and automation. From manufacturing to automotive and marketing to sales, artificial intelligence has positively impacted almost every domain. AI and ML (Machine Learning) algorithms have helped automate numerous tasks that previously needed human intervention. Robots are a particularly good example of this as they automate tasks across the agricultural, industrial, commercial, and medical sectors.  By leveraging vision and AI, robots enabling human workers to focus more on strategic activities.

Even though the integration of AI and ML into robots has dawned a new era where autonomous navigation and intelligent image analysis have become a reality, developing a robotic system from scratch can be very challenging. A robot is a complex system requiring many elements to work together in tandem including processing, sensors, and communications systems.  Even more importantly, the system must perform its duties safely, and when something fails, it must fail in a safe way.

The way that many robotic systems are designed today makes use of off-the-shelf components including IPCs, wireless routers, safety PLCs, and GPUs.  The value added by most robotic platforms is in the application of the system, the hardware platform, and the software running the platform.

This is why we see that there is an alternative to the use of expensive GPU and off-the-shelf components that are not ideally suited to robotic systems. In this article, we explore how TechNexion’s ROVY-4VM is the perfect system on module for edge AI-based autonomous robots. You will learn how most robotic systems are architected today, their limitations, and how our solution can help you overcome them.

The AI-enabled robotic revolution

Autonomous robots are not the future anymore. AI-enabled robots are already aiding in tasks like automated weed removal, sea floor mapping, order picking, inspection, and more. Here, let us look at the various ways in which robots are augmenting – and often eliminating – human efforts across various applications.

Agricultural robots

According to the USDA, production expenses in agriculture continue to rise. Just from 2021 to 2022, USDA estimates total production expenses to increase 5%; that’s after a 9% increase from 2020 to 2021. This reiterates the need for further optimizing agricultural practices for higher efficiency and lower costs. Deploying robots is one of the ways to do this.

Here are some ways in which robots are enhancing agricultural operations:

  • Automated weeding: Weed removal robots use a combination of cameras and AI/ML algorithms to kill non-beneficial plants. For example, Carbon Robotics uses a high-energy laser to kill weeds in real time with high precision.
  • Automated picking and delivery: Robots can automatically pick and deliver produce from the fields to the processing facility or any desired location. The ability to precisely measure depth is a key feature required for robots used for this purpose.
  • Spreading fertilizer: Robots are used to spread fertilizer based on aerial multi-spectral data obtained from drones. By analyzing the aerial image data, the stage of growth can be identified, which can then be used as input for robots to apply fertilizer to certain sections of the field.
  • Plowing: Automating plowing using robots saves a lot of time and costs for farmers. They no longer have to manually run the tractor through large fields, which leads to significant savings in terms of labor, fuel, and maintenance costs.
  • Pollination: Given that the bee population has declined by 7 percent from 2022 to 2023 in the US alone, robots have emerged as an alternative to ensure sufficient pollination. Robots can also improve the effectiveness of the process by using artificial intelligence to identify the optimal time for pollination.

Maritime robots

The application of robotics in sea exploration and related activities is gaining more attention owing to the convenience it brings. In addition, given the challenges of exploring the ocean floor, maritime robots offer a better and more effective alternative to human-based explorations.

Camera-enabled maritime robots can be used for:

  • Mapping out the ocean floor with unmanned boats and cruises.
  • Identifying and destroying harmful sea weeds.
  • Aiding in finding lost and submerged ships. Robots can also help with the repair process of submerged equipment like anchoring cables and scientific research devices.

The powerful combination of AI and cameras in maritime robots extends the possibility of what oceanic exploration can do. From navigating tough terrains to accessing challenging locations and analyzing oceanic life, robots offer a shiny future for maritime applications.

Warehouse robots

According to the study Applications of Industry 4.0 Technologies in Warehouse Management, some of the benefits warehouses receive from Industry 4.0 technologies include:

  • Improved process efficiency
  • Availability of real-time data
  • A competitive advantage for a business
  • Digitization
  • Enhanced flexibility
  • Improved decision making
  • Reduced operational costs
  • Better risk management
  • Enhanced security

Robots are a key component of the industry 4.0 revolution. From picking & placing and automated packaging to automated dimensioning and loading & unloading, a wide variety of warehouse tasks can be automated using robots such as:

  • Goods-to-person robots
  • Pick and place robots
  • Automated forklifts
  • Robotic arms

Industrial inspection robots

Inspection robots are used for inspecting damage and corrosion in industrial equipment and surfaces. Instead of manually measuring the width of a crack or the extent of corrosion, robots can use image recognition algorithms to quantify the damage. Given that they can operate 24×7, the dependency on human labor is also reduced. These robots are critical in ensuring industrial setups run without interruption by regularly conducting inspections.

Construction robots

According to Construction Connect, a recent AGC (Associated General Contractors) workforce survey shows that “88% of construction firms are having a hard time finding workers to hire.” This has accelerated the adoption of robotics in the construction sector with these smart devices being used for the following tasks:

  • Automated painting: Robots equipped with 2D and 3D cameras can reduce wastage by optimizing paint quantity. This is achieved by analyzing surface irregularities on the go.
  • Drywall finishing: AI-enabled robots can tape and apply finishing compounds onto dry walls. This significantly enhances the safety of construction sites by eliminating potential human injuries during drywall-related works.
  • Heat welding: Robotic heat welders are also great examples of enhancing safety by easing the welding process.
  • Demolition: Robots can also help with the demolition of buildings and structures by autonomously navigating the site.
  • Surveillance: Job site monitoring and inspection activities can be automated to a large extent with the help of robots. Whether it is ensuring that various equipment is in place or monitoring restricted access, robots can be extremely useful in monitoring construction sites.

Drones and UAVs

Drones and UAVs (Unmanned Aerial Vehicles) are making logistics, imagery, and inspection easier across the industrial, commercial, and agricultural sectors. With the help of cameras, drones can fly from one point to another autonomously.

The current experiments with autonomous drone delivery are a great step in the field of UAVs. Some of the other applications of drones include:

  • Remote inspection in industrial and construction sites.
  • Smart surveillance of public places, industrial sites, and commercial settings.
  • Aerial imaging for agriculture.
  • Nature and oceanic exploration.

The architecture of robotic systems

Now that we understand how the vision and AI-led revolution of robots has been impacting multiple industries, let us dive deep into the different components and systems that make a robot. This is key in establishing why a different approach might be needed to equip your robots for future-ready applications.

Industrial PCs

Despite the wider adoption of on-device processing systems, many robots today leverage high-end industrial PCs (IPCs) to perform the main compute tasks. These include sensing, vision processing, and control loop functions. Most robotic systems on IPCs are run on ROS (Robotic Operating System), which offers a structured module for developing robotic software.

Below are some of the key reasons why many robotic product developers go with IPCs:

  • Industrial PCs are easy to work on with familiar software. They are well-suited for meeting any kind of application dependencies.
  • They are easily scalable in terms of CPU throughput and memory. This helps to address a wide range of processing needs.
  • The IPC market is well-established, and hence robot manufacturers get to choose from a diverse set of IPC vendors.

Despite the wider adoption of on-device processing systems, many robots today leverage high-end industrial PCs (IPCs) to perform the main compute tasks. These include sensing, vision processing, and control loop functions. Most robotic systems on IPCs are run on ROS (Robotic Operating System), which offers a structured module for developing robotic software.

Below are some of the key reasons why many robotic product developers go with IPCs:

  • Industrial PCs are easy to work on with familiar software. They are well-suited for meeting any kind of application dependencies.
  • They are easily scalable in terms of CPU throughput and memory. This helps to address a wide range of processing needs.
  • The IPC market is well-established, and hence robot manufacturers get to choose from a diverse set of IPC vendors.

Though IPCs come with these advantages, they tend to be on the costlier side, especially if you want advanced features such as:

  • A wide variety of embedded interfaces: In addition to USB and PCIe, many lower cost sensors and actuators communicate with simpler interfaces such as I2C, SPI, UARTs (serial ports), and CAN bus
  • High-speed native video interfaces: MIPI-CSI2, the commonly used interface with the lowest-latency and highest fidelity is not a common feature supported in IPC chipsets.
  • Extended temperature ranges: Achieving extremely high temperature resistance requires high-end design and components, which in turn increases the cost.
  • High shock and vibration resistance: Many IPCs utilize a SO-DIMM memory module, which is more susceptible to problems during mechanical vibration.
  • Accelerators for video processing and edge AI: Most IPCs would require an accelerator or GPU to achieve good video processing and edge AI performance
  • Networking: Many robotic systems resemble in-vehicle networks (IVNs), with sensors and actuators connected using Ethernet and IP-based protocols.
  • Real-time Control: IPCs are comprised of multicore processors from AMD or Intel, which are generally running a single OS such as Linux, with limited real-time processing capability.
  • Functional Safety Support: As robotic systems become mainstream, adding a safety processor to an IPC will be a necessity.

Sensors

There are two primary sensor domains in robotics.  Sensors that allow the robot to understand the world around it (external), and also sensors allowing the robot to monitor its own health and status (internal). To humans, external sensors are like our external senses (e.g. sight, touch, sound, smell, and taste) and internal sensors are analogous to our own internal body senses (for example pain, hunger, joint position, muscle engagement, and fatigue/tiredness).

External Sensors

2D cameras: Robots use 2D cameras to detect obstacles, perform surround view, track inventory, and perform all imaging-related tasks.  With more advance vision processing capabilities, we will see significantly higher numbers of 2D cameras in use for a variety of tasks.

  • 3D depth cameras: Depth cameras such as stereo cameras, structured light cameras, and ToF (Time of Flight) cameras help robots to accurately measure depth for precise picking, placing, harvesting, weed removal, and similar tasks that require distance calculation.
  • Proximity sensors: As the term suggests, these sensors are used to detect the presence of any object or obstacle near the robot.
  • Radar: Radar use transmitted RF energy to determine the distance to objects in a robot’s environment. While it has the disadvantage of being lower resolution than light-detecting sensors such as cameras and LiDAR, there are numerous advantages including ability to perceive in foggy and dark environments, detect minute movements, and even in some cases even see through walls.
  • LiDAR: LiDAR is a type of ranging technology that uses lasers to measure depth to long distances. The maximum range of a long-range LiDAR can go up to 2000 meters.
  • GPS: In spaces where a view of the sky is solid, the GPS can be used for locating the position of a robot.

Accelerators

When we refer to accelerators in robotic platforms, we usually refer to processing systems that accelerate and offload certain key tasks that require a very high computational load.  In addition, usually this type of computation does not run very efficiently on general purpose CPU cores.  Dedicated and optimized accelerators are much more suited to these tasks.

In general, we are talking about the following types of accelerators:

  1. Video accelerators: These accelerators offload video processing workloads from the main CPU, usually for the purpose of making a video stream suitable for further processing. Examples of video processing are, image signal processing, colorspace conversion, video format conversion, video compression, and image scaling.
  2. Vision accelerators: Vision accelerators are involved in perception tasks, taking 2D image data and converting it into useful information.  For example, taking video input from two image sensors in a stereo configuration and converting that into a stereo depth map (or point cloud), or performing optical flow measurements on it (for odometry purposes).
  3. AI/ML accelerators: AI/ML accelerators offload the CPU from AI inference processing tasks (running AI/ML models), which are very useful in perception applications including object detection and monocular depth estimation. These tasks require tremendous amounts of fixed or floating-point matrix math operations and are much more efficiently run in dedicated hardware.

Many IPC-based systems will not have embedded hardware for video processing acceleration and instead they rely on the CPU or GPU for video acceleration. This is not an efficient method since it draws more power and CPU bandwidth. When it comes to AI/ML acceleration, GPU resources are required in an IPC-based system.

Wireless radios

Robots use different radio techniques for short-range and long-range communication. While WLAN (Wireless Local Area Network) is used for on-premise communication, 5G WWAN (Wireless Wide Area Network) is used for long-range communication, and in some cases is being used in private networking applications.  5G WWAN is also used for data backup to preserve critical data. On the other hand, for close-range communication, Bluetooth is the preferred choice.

Networking

IP-based networking, including Ethernet, is heavily used in robotics.  Robot Operating System (ROS) can use IP-based transport to communicate between nodes.  Many sensors, such as LIDARs and Ethernet cameras are natively IP-based sensors because of the high bandwidth and ease of software and hardware integration.  Ethernet can also be used to connect other components such as actuators and drive controllers to the main control processor.

Safety Controllers

Most mobile robots and collaborative robots (Cobots) are operating in environments where a failure of one or more subsystems can mean a significant cost.  For example, if an AMR, delivering a hefty load of goods fails to detect a shelving rack or, even worse, a person, there may be damage to the robot, loss of operating income, or even risk of injury or death.   Safety controllers, which perform safety functions such as failure monitoring and take action to minimize the consequence of failure.  For example, an safety processor within an AMR might monitor proximity sensors or LIDAR and take action to prevent collision, such as turning off power to the drive motors.  It is important that this function is performed completely independently of the main controlling processor.  This is called functional safety.

Real Time Control

When we discussed sensors above, listed out two different categories of sensors – internal sensors (used to understand the internal state of the robot) and external sensors (used to understand the outside word).

This is a pretty simplistic view of the sensor domains.  In reality, just as in our bodies utilize both internal and external sensors to interact with our world, so must robots.  One only has to try picking up a glass of water from to see how your senses of sight and touch must coordinate with your internal senses of muscle exertion and joint position must all be used to effectively perform even the simplest tasks.

IPC-based ARM System

The limitations of using a high-end processor in robots

High-end processing systems like NVIDIA Jetson Orin offer a lot of video and AI bandwidth and compensate for the loss of processing power due to an inefficient software. However, they come with some disadvantages such as:

  • High costs that might put a dent in your budget.
  • May require a lot of integration and I/O expansion.
  • The processing power is not necessary for most workloads.

In addition, safety is also a huge concern here. Given that humans work and interact closely with robots, heavy processing systems can lead to an increased risk of accidents and damage. For example, what happens if the motors are commanded to move forward immediately before an unplanned reboot or shutdown of the main computer?

The reality is that safety requirements are sometimes only a consideration after the platform has been designed to fulfil its primary mission. This can result in expensive additional components being added such as safety-certified PLCs. Additional components and integration efforts like this add to the overall cost of the product development process, which may even lead to compromising some of the key features of the system to limit the total budget.

Given this, the key challenge facing product developers is that robotic systems aren’t scalable unless you have a large budget and extensive engineering expertise. In the next section, we look at how an alternative approach can solve these challenges in addition to learning how TechNexion’s solution is a perfect fit for modern-day robots.

Designing robots of tomorrow with TechNexion’s ROVY-4VM

Many commercial autonomous robots are designed with a lot of flexibility by keeping expansion needs in mind.  Often, there is a lot of inertia associated with existing design components. As robot developers add more features, the design gets even more complex, making it difficult to meet the ever-evolving capability requirements of new-age robotic systems.

The robots of tomorrow must be built with scalability in mind while balancing costs. This is where TechNexion’s ROVY-4VM becomes a game-changer. Based on the TI TDA4VM Jacinto™ processor, it is a SoM (System on Module) that has all you need when it comes to building a futuristic robot. Let’s dive into how the solution helps overcome some of the burning problems facing robotic product developers today.

ROVY-based ARM System

In pursuit of building a scalable solution

TechNexion has been one of the frontrunners of innovation in the embedded systems world. We have always been in search of finding the right components to build solutions that meet the end application needs while ensuring they fit our customers’ pockets. In an effort to build a specialized SoM for robotic systems, we evaluated multiple platforms. The platform had to come with:

  • All of the elements required for robotics compute workloads. These include vision capability and the ability to capture and process different video feeds. It should also have onboard AI acceleration for running complex AI algorithms.
  • The right kind of interfaces needed for robotic systems today as well as in the future. While the present is about having lots of CANbus, serial ports, SPI, I2C, and Ethernet, robots of tomorrow will need added Ethernet ports as more sensors become IP-connected. In addition, connectivity for different radio communication protocols like WLAN and 5G was also required.
  • A functional safety-based design.
  • Fit for long-term use in challenging industrial and outdoor environments.

After evaluating multiple platforms, we found TI’s Jacinto series of SoCs (System on Chips) to be the best-suited solution for robotic systems.

Architecture of TI Jacinto TDA4VM (Source: TI)

This series of processing systems is designed for automotive, ADAS, and similar applications with features such as:

  • Wide temperature operation
  • Rugged industrial design
  • ADAS processors with built-in vision capabilities
  • Many communications interfaces
  • Functional safety certifiable

Understanding TechNexion ROVY-4VM: TI TDA4VM System on Module

ROVY-4VM is a rugged System on Module that is specially designed for autonomous vehicles like robots. It leverages the key features of the TI TDA4VM platform with a modular approach – allowing us to offer this core compute engine at the least cost while giving you the flexibility to make your own design decisions.

ROVY-4VM: TI TDA4VM System on Module

ROVY-4VM is forward compatible with TechNexion’s upcoming products in the series such as the ROVY-4VH. It can also be the basis for system level industrial controller products.

In addition to these, given below are some of the standout features of the SoM:

  • Comes with all the necessary accelerators and integrations such as vision, AI, I/O, and functional safety. With a performance of 8 TOPS, it supports up to 10 CAN, 9 GbE, 4 PCIe, and 8 cameras (via MIPI virtual channels). Also, with its rugged design, it can withstand extremely high temperatures in industrial and commercial settings.
  • ROVY-4VM offers you the flexibility to design your own robotic controller or use ours for your product development.
  • It comes with a ready-to-go software platform based on Yocto. It can be customized based on the end application requirements. We also have the Debian software planned in the product pipeline, which is expected to be released soon.

How TechNexion’s TDA4VM SoM compares with other solutions

ROVY-4VM is a ‘balanced solution’ that comes with all the necessary accelerators, I/O configurations, and rugged design that will make your robotic product development faster and easier, thereby enabling you to hit the market faster. A faster time to market in turn means a faster revenue realization, leading to speedy growth.

But how does ROVY fare compared to other solutions in the market? Let us discover in this section.

NVIDIA

One of the biggest disadvantages of NVIDIA-based solutions is that they are expensive, making it often impractical for early-stage (or even mature) robotics companies. ROVY is a more affordable alternative with better I/O features. It also comes with high temperature resistance the ability to work in the I-temp range. ROVY is also FuSa (Functional Safety) certifiable.

Industrial PCs

ROVY is a more affordable option compared to industrial PCs. It also offers better integration options in addition to having built-in AI and vision processing capabilities. The FuSa compatibility also adds to the list of key differentiators of ROVY.

Qualcomm

Qualcomm offers high-quality products. At the same time, their SoCs do not have network interfaces onboard. Instead, they rely on external components for this. Moreover, we believe ROVY is a far more cost-effective solution in comparison with Qualcomm-based products.

The ROVY Family

Conclusion

One of TechNexion’s key objectives is to make product development easier for robotics companies. We believe that robotics has a key role to play in building a smarter and more intelligent world, and we wish to be a part of that revolution by enabling you to take your robots to the market faster, that too at a lower cost.

In pursuit of this, we have heavily invested in the research and development of comprehensive robotic solutions including SoMs and embedded cameras. With new products such as ROVY-4VH in the pipeline, we aim to consistently add value to our customers’ lives by building new-age products that will become cornerstones of futuristic robots.

Visit the ROVY product page to learn more about the product. To discuss how the solution can accelerate your robot development journey, get in touch with us here.

Get a Quote

Fill out the details below and one of our representatives will contact you shortly.

SHARE YOUR CART