Network Considerations for Advanced Robotics

In 2000, Honda introduced ASIMO (Advanced Step in Innovative Mobility). At the time, this was a truly pioneering piece of engineering, a humanoid robot capable of walking and human interaction. 

ASIMO itself had been in development for almost three decades, with the first (sort of) walking prototype achieved in 1986. And yes, when ASIMO was shown to the world it was slow and its movements were jerky, and (like a Dalek) it couldn’t climb stairs.

But there is a reason why so few creatures walk on two legs, and why humans take so long to walk versus four legged animals. For humanoids (both actual humans and robots), the computational power, the processing ability, and the signal transmission are all (to understate significantly) incredibly difficult to execute.

Since ASIMO, engineers have come a long way. Recent advances in AI, computational power, sensing technologies and data transmission have seen robots become more agile and capable. Today’s humanoid robots can simulate speech and hold conversations, navigate rugged terrain and even mimic highly dextrous human skills like playing the cello (see our white paper on the technologies underpinning these advances). Indeed, earlier this year advances again made international news when robots danced alongside humans and performed Kung Fu routines in live demonstrations for the Chinese Lunar New Year.

And behind every precise movement, there’s a high-speed, distributed control system with a brain that delegates tasks to a network of localized embedded microcontrollers to manage a huge number of motors, sensors, and actuators. For the system to work seamlessly, though, the network must handle large data volumes with minimal latency and jitter.

Comparing Industrial Communication Standards for Humanoid Robotics

In this blog, we'll compare and contrast four major transmission standards: CANopen, Ethernet, PROFINET and EtherCAT^® to create an interconnect strategy for such humanoid robotic systems.

CANopen

Originally developed for time-critical vehicle functions such as steering and braking, CANopen is a standardized communication protocol used in automated systems to enable different devices and nodes to exchange data efficiently. It also provides standardized methods to configure and customize those devices, making it highly flexible across different types of hardware.

The standard includes bit error detection and is exceptionally robust and is highly trusted by virtually all automakers, as well as by industrial machine tool manufacturers. CANopen also supports the cable lengths needed for humanoids and larger robots, and is a highly cost-effective network architecture for real-time control applications.

CANopen works through a publish-subscribe model, whereby a producer device publishes data onto the CAN bus and any number of other devices (message recipients) can subscribe to receive messages in those frames. This approach allows a single device to deliver data to more than 100 nodes using the same data frame, enabling a great degree of synchronization.

Ethernet

While CANopen uses a linear spine where all nodes communicate over the same backbone, Ethernet networks are typically arranged in a star or tree topology with a central switch. This switch forwards messages only to the intended device, with data rates that can scale up to hundreds of gigabits per second.

However, despite its higher bandwidth, traditional Ethernet does not guarantee the deterministic delivery of high-priority messages, which is critical for robotic motion control.

One of its core limitations comes at the medium access level where, because the network is shared, any devices trying to send data at the same time as another could potentially cause packet collisions. When this happens, the Ethernet network interface would halt transmission and wait for a random interval before resending the packet. This variability makes it difficult to guarantee that jitter stays below the hundreds of nanoseconds threshold that is required by modern robots.

At the network level, some Ethernet variants use the internet protocol to enable devices to communicate with each other across networks. The benefit of using IP is that it makes sure every packet will be sent to the right address, but IP doesn’t have a built-in mechanism to keep network delays low or predictable, again making it less well suited to robotics.

PROFINET

PROFINET is an evolution on the PROFIBUS fieldbus system used in factories, using Ethernet to deliver increased transmission speeds and bandwidth, but with the PROFINET Real-Time (RT) and Isochronous Real-Time (IRT) features that ensure data gets delivered on time and reliably.

In Profinet RT, data packets are classified by importance, with the system prioritizing the transmission of critical data to ensure it gets through, even when the network is heavily loaded. This feature also splits lower-priority messages into smaller pieces, so they can be sent without getting in the way of more important traffic. However, it should be highlighted that PROFINET RT still comes with a risk of unacceptable jitter during periods of network congestion.

PROFINET IRT handles these timing issues by setting aside dedicated time slots for important data that needs to be delivered right on schedule. For example, if network traffic is divided into five time segments, four would be used for regular RT communication, while one segment would be reserved for IRT communication. In this case, the network would dedicate 20% of its bandwidth to PROFINET IRT and the reliable transmission of critical data packets.

This evolution provides many benefits over Ethernet, but the main network still relies on standard Ethernet and both cycle times and jitter cannot be fully eliminated.

EtherCAT^®

EtherCAT^® was designed specifically to handle real-time control, making it a great fit for humanoid robotics and other applications that need precision timing.  

Unlike standard Ethernet, where devices wait for a full packet to arrive, EtherCAT^® processes data while still in transit, with a device reading data as it passes through a node and adding its response into the same frame before it exits the node.

To prevent data collisions, the standard has one main device manage all network traffic, and implements a distributed clock that keeps all nodes tightly synchronized. This reduces jitter to about 100 nanoseconds, which is well within the threshold for humanoid robotics. This gives the same elite precision as PROFINET IRT, but without the requirement for (and cost of) specialized hardware in the main device.

The standard also gives a level of flexibility not present for the above in relation to wiring, with it being possible to wire it like a tree, a star, or as a redundant ring for instant fault recovery. And if a cable segment fails, the network automatically reroutes traffic to maintain communication, which eliminates interruption in critical operations.

Communication Standard Comparison Summary

Integration at the MCU Level

To help make this technology easier to use and develop, GigaDevice has built an EtherCAT^® SubDevice Controller into its GD32 high-performance MCUs. For example, the GD32H75E combines a fast Arm^® Cortex^®-M7 core with a dual-PHY EtherCAT® controller, Ethernet ports and three CAN FD interfaces. It also comes with motion control tools, such as encoder interfaces and hardware that can handle digital filtering faster.

By combining all these features into one chip, designers can save valuable space on the circuit board, which is especially important for building compact robotic limbs, and achieving the tight integration needed for real-time closed-loop motor control.

For more information, please read our EtherCAT^® blog or visit our humanoid robotics page. 

Or, for a more in-depth analysis, please see our original article published on Embedded.com: Selecting the best network for robot control.