Distributed Sensors for Autonomous Robots

Science fiction has, for generations, integrated robots into society in a variety of ways, from helpful, benign servants that make our lives easier and better to harsh, soulless overlords trying to exterminate us. In all cases, the story begins with robots replacing people to complete menial, repetitive, and dangerous tasks, and with AI's emergence, they outperform even highly trained humans. 

It is not uncommon to see them in use, from little floor cleaners to delivery robots and drones. Society is ready to integrate them into our lives, especially as they become autonomous, conversant, and even self-aware. 

For decades, workhorse robots in car fabrication plants, chemical plants, and factories have replaced many humans and shown cost-effectiveness and higher productivity and accuracy, as manufacturing and industrial applications have been the driving force behind robotic advancements.  
While initial investments are still high to purchase, program, and put manufacturing robots to work, they bring many benefits to manufacturing companies. 

With emphasized desire to bring manufacturing back home, competitive pressures mean that only robots can maintain high productivity at low cost. They can work twenty-four hours a day without breaks. They don’t take vacations or holidays. They don’t call in sick (but they do break down every so often). 

The key is that manufacturing companies that employ robots don’t have to pay social security, unemployment, workman's compensation, medical insurance, retirement contributions, etc. 
As fixed location and function robots are replaced with mobile, autonomous, multifunctional learning machines, capability, safety, and cost sensitivities are more important. Robot designers now face choices regarding architecting their robots and their robot’s working environment. 

Ten Kilos of Chips in a five Kilo Robot

Sensors are key in any robot design. They provide machines with information about the outside world. While older generation fixed robots don’t need to navigate, sense obstacles, know where they are, and find alternate routes, new ones do. Thus, many types of sensors must be implemented, monitored, prioritized, interpreted, and used to determine actions. The question facing designers today is how to distribute sensors, while maintaining a reasonable size and weight. Is everything placed inside the robot? Or does the robot get sensory data and instructions from a facility, wireless controller, or cloud?

The more you make the robot do, the larger it needs to be. The heavier the robot, the more battery power it requires. This limits the run time, speed, and performance and increases cost due to taking it out of service while it recharges or battery swaps. 

The space is limited, and weight is the enemy. As a result, the more intelligence you place in the environment, the less needs to be embedded in the robot. Smaller, faster, more runtime, lower cost. Win-win, right? Well, only so far. 

Even with an intelligent environment, smarts and sensors must be embedded into each robot. Fault tolerance means that safety isn't compromised in the event of a communications failure. So, the use of lightweight onboard sensors like transducers, video cameras, temperature and battery voltage sensors, etc., are feasible. Even 2D Lidar technologies available today can be integrated into a low-cost, efficient robot. 

Failsafe sensors like proximity or power issues should be prioritized. For example, a robot may not know it is being held in place by an obstacle, but a current sensor on the drive motors can detect a fault quickly and remove power to the motor drive immediately. A central facility computer or cloud controller can't react quickly enough to prevent a possible fire.  

In a factory or industrial setting, wireless communications can be a challenge. A manufacturing floor is an electronically noisy environment. If errors and retries occur, a robot may not get the signals needed in time to preserve safety. It also may mean that wireless links transmit higher power to obtain reliable communications.  

Open World, Closed Floor; Distributed Communications Options

A unique approach that can solve many of these issues is using a closed, distributed network for sensory array deployment and back-channel information from the robots. Using communications nodes scattered around a manufacturing floor can create a closed-loop feedback mechanism where the facility controller gets high-level verification for every step a robot takes. 

Sensors and communications nodes can be strategically placed at specific and planned locations. A 5G, Wi-Fi®, or Bluetooth® signal becomes much more reliable at ten feet than at fifty feet. So right away, a reliable primary and/or backup communications scheme can be realized at a low cost. Even the lowest cost narrow band AM or FM data links can operate concurrently with high-level communications. For example, a mobile robot passing a stationary robot can gather its safety data and pass it along. 

Robots passing these digital checkpoints can report on battery level, internal temperatures, vibration levels, stress, and strain on loads they may be carrying, as well as the health of their central processor, vision systems, and overall status. Radio Frequency Identification (RFID) technology can even be used to verify robots as they pass through single file checkpoints. As you can see, even with a high level of autonomy in this approach, the robot becomes like a peripheral to the manufacturing floor. 

Distributed Serial Architectures: Topologies and Benefits

Serial communications are ideal for distributed communications around a localized or vast area. System and facility designers and managers have several choices and options for the technology and types of serial networks that can easily be implemented. 

Ethernet is a serial network technology that can be used. It can be wired using Cat5 and/or Cat 6 styles of multi-twisted pairs in a single cable. Eight wires are needed for each point-to-point Ethernet connection which increases costs and decreases reliability. Ethernet does offer fast communications rates and the ability to run a few hundred feet and can be used as hubs for serial networks in zones. Still, it is not particularly noise or Electro-Static Discharge (ESD) immune, so it is not a good choice for machine locations. In addition, each node on an Ethernet network requires more processor resources and bandwidth. In contrast, a lower-cost Universal Asynchronous Receiver/Transmitter (UART) framed asynchronous packet framed serial data link can be used. 

UART-based serial links are perhaps the most straightforward and most reliable. Wireless serial links are rugged and reliable even in the highest Electro-Magnetic Interference (EMI) noise and harshest ESD environment. There are also a variety of drivers and receivers that can go hundreds and even thousands of feet, maintaining a range of data rates. 

Architecturally, a distributed serial sensor array can be controlled and monitored by a central location, distributed, or remotely accessed. These are not mutually exclusive. For example, a central computer can always run the show, while a backup system monitors and kicks in if there is a fault with the primary system. 

There is also point-to-point flexibility built into serial networks. The simplest and most reliable architecture is point-to-point direct (Figure 1A). This is the fastest, most direct link. Hardware and/or software handshaking can assure reliable data transfer. 

A string of daisy-chained devices can be configured where a single data link communicates with many distributed sensors or actuators (Figure 1B). Data passes through each node and routes back to the initiation point thru line receivers and drivers that act as pass-through buffers. Each node can be individually addressed, or all accessed simultaneously as would be the case with an 'Emergency Shutdown" command, for example. 

If devices are configured in a loop, then the pass-through drivers can be eliminated and all data returns to the initiation point. Data rates are typically set to a single baud rate for a string or loop architected network. A Tree configuration is also possible if several multiport serial console controllers are used (Figure 1C). This allows multiple sensors or sensor zones to be established in the serial domain if data rates from the host are fast enough to supply all lower levels. If not, Ethernet connectivity can be used as the top-level console server. This can be a helpful architecture if sensor zones are defined with a localized area.

Figure 1: (A) Point-to-Point Direct connect. (B) Loop or Daisy Chain of linked devices. (C) Tree configuration. (Source: Mouser Electronics)

In addition to remotely distributing sensors and actuators, distributed nodes can include dedicated RF links for safety control or RFID scanners to read and write data to robots in the facility as they pass specific key locations. Any time a safety redundancy can be engineered at virtually no extra cost, it is wise to take advantage of that safety redundancy. 

Types of Serial Link Layer Protocols

An oldie but goodie most of us are familiar with is RS-232. This single-ended physical layer scheme uses a non-return to zero (NRZ) signaling protocol that represents a logic zero as a positive voltage and a logic one as a negative voltage. The signal line is never resting on the ground. As a result, RS-232 is relatively robust and rejects noise and impulse glitches well. Fault detection is easier since an unconnected wire will not show voltage. 

While RS-232 can traverse a few hundred feet, RS-422 can traverse a few thousand feet. RS-422 uses a differential pair of wires for each signal. This makes it immune to common-mode noise that can be present in a noisy factory setting. It requires twice the number of wires and is not as common as RS-232, which was used for most computer peripherals and modems of the past. 

RS-485 is also a differential medium but is also a shared medium. All RS-485 devices connect to the same physical wire pair. This means that RS-485 devices can only operate in half-duplex mode, whereas RS-232 and RS-422 can operate in a full-duplex mode. All RS-485 devices typically have the same communications characteristics such as baud rate, parity, and stop bits. 

RS-485 can also span several thousand feet when appropriately implemented and is used commonly in the factory and industrial settings. In all cases, semiconductor makers provide cost-effective, well-engineered, competitive line drivers and receivers for all of these, with ESD and noise protection built in. 
Note, UART framed asynchronous packet formats like RS-232, RS-422, and RS-485 also are used in the automotive world for the Controller Area Network (CAN) and OBDII control and sensor busses. 

The older but popular Midi interface also used a UART-based packet protocol. These signaling techniques are well understood and supported with tools and development firmware. They are ideal for distributed sensors and actuators in an industrial setting, especially if a multiport server can be used.  

Medial Server Solutions

Multiport consoles provide multiple individual serial ports accessible through IP, such as Ethernet and TCP/IP access points. Available in 8 or 16-port RS-232 configurations, the ports can be used as RS-485 or RS-422 with the appropriate interfaces. 

The two 10/100/1000 Ethernet ports can support the total bandwidth of up to 921.6Kbits/sec data rates of the individual serial ports and allow connectivity to local computer resources as well as remote and distributed global computer and cloud-based services. 

The standard 1U rack mount configuration allows easy integration into virtually any facility. Additionally, each serial port protection scheme has built-in 15KV ESD protection. These console servers act as the ideal interface between TCP/IP networks and PLCs, CNC machines, scales, scanners, and virtually any type of sensor-based systems. Many sensor systems that can be Original Equipment Manufacturer (OEM)’d for easy use support these serial standards, and the console servers allow individual connectivity to these sensors. Note that each one can have its baud rate, parity, and stop bit configurations simplifying setup and use and maximizing the bandwidth since the slowest device doesn't limit the faster ones. 

Control through Windows utilities, web browsers, Telenet, and various consoles allows operators and programmers an easy setup, operation, and monitoring. Network management tools like SNMP MIB-II also allow standard toolsets for operation and monitoring. Rear-mounted wiring keeps implementations clean and serviceable for easy expansion or updates. Various topologies can be implemented using the built-in Ethernet connectivity to create a simple, low-cost, effective distributed sensor array. 

Example Factory and Distributed Robot Sensors:  
For example, we can visualize a central computer system as the master, which connects to serial console servers through an Ethernet switch. This provides the high-speed bandwidth to each zone where the console controller acts as the interconnect to the individual sensors. 

In Figure 2, an example topology uses a redundant control link through higher-speed Ethernet switches to route to localized zones. Each localized zone can take advantage of the serial console servers to establish direct connect links to every sensor, RF link, or actuator our robots will be passing by. 

A remote or redundant link can be established as a backup, watchdog, or failsafe even if a primary controller fault pops up. This can be a local redundant controller or a remote or cloud-based monitoring and control system.

 
Figure 2: Example topology (Source: Mouser Electronics)

Conclusions

Combining embedded and distributed sensors in a robot's environment can be the safest and most reliable way to implement a factory or industrial setting that takes automation to the next level. Zones can be set up, and redundant safety mechanisms can be monitored to ensure that all systems are functioning correctly. 

Serial ports are still the most cost-effective way to distribute sensors and actuators.
Virtually every microcontroller has UART-based communications, and OEM and designed sensors can easily integrate into this type of network. This also allows robots to be streamlined and less complicated since many of their responsibilities can be offloaded to a more intelligent facility.