Bi-directional learning: The future of robotics

Machines are now being designed to incorporate bi-directional learning: the data sensed by an individual robot is sent for processing to determine an appropriate action. The revised instruction is then disseminated back to the machine.

June 26, 2020

5 Min Read

Robotic technology has revolutionized global industry; machines have been replacing humans at a phenomenal rate, given their suitability for completing repetitive tasks, day in, day out, without tiring or losing focus. Now, as the COVID-19 pandemic is accelerating a more digital approach to accommodate social distancing measures, many businesses are rekindling plans to invest in automated technology to increase productivity and reduce overheads.

Yet before taking the next step, companies should consider the potential limitations of ‘non-learning’ robots, such as those commonly deployed on manufacturing assembly lines. Each automated device is a link in a greater chain, undertaking tasks – such as loading or assembling individual items – with the necessary cooperation of co-worker robots. Unlike humans, ‘non-learning’ robots aren’t adaptable or able to deviate from their pre-programmed tasks. Which means that any unexpected events can cause issues with a robot’s ability to sense or relay information back to a diagnostics center capable of receiving new instructions via real-time software updates. Thus small problems quickly escalate to bigger failures across the whole production line. A solution requires manual software updates and regular re-configuration to ensure that these robots continue to work harmoniously with other machines within the production space.

However, as technologies rapidly evolve, more complex machines are being developed with applied sensor-based technology and artificial intelligence (AI) capabilities. These advancements allow a robot to interpret its surroundings and adapt to changes in its assignment in real-time. In addition, machines are now being designed to incorporate bi-directional learning. The data sensed by an individual robot in a situational ‘learning’ scenario is sent to a core algorithm for processing to determine an appropriate action. The revised instruction is then disseminated back to the machine.

This represents a shift towards a more contextually aware system of robotics, which requires interconnection not only to other robots, but also to the internet of things (IoT) sensors and other digital ecosystems to support AI data processing and model building. This enables each machine to harness data and processing power to detect and anticipate changing circumstances in the environment and identify appropriate responses. When current, sensed data is combined with historic data from past incidents, issues are identified and eliminated before they develop. As available data increases over time, the AI system evolves for even more accuracy in decision making.

The ability to process the huge amounts of data required to create ‘intelligent’ robots is dependent on a distributed, hybrid multi-cloud IT infrastructure with low-latency and secure connectivity for private data exchange. Dubbed cloud robotics, this setup provides almost limitless processing power and storage resources to expand machine capabilities. Robots are able to offload all data intensive tasks to the cloud, including image recognition, mapping systems, machine learning and software updates. This convergence of IoT and AI domains and their supporting IT architectures can infuse greater intelligence into frontline robots on the factory floor. Equinix’s own data center platform, ECX Fabric, facilitates this capability by enabling companies to rapidly and securely move and process vast amounts of data in the cloud.

It’s a concept that has evolved over the last five years. In 2014, ABI Research coined the term the ‘Internet of Robotic Things (IoRT)’ to describe the system, claiming that “intelligent devices can monitor events, fuse sensor data from a variety of sources, use local and distributed intelligence to determine a best course of action, and then act to control or manipulate objects in the physical world.” In contrast to pre-programmed industrial robot, IoRT systems allow robots to exchange data and insights to adapt to changing situations in real-time.

This bi-directional learning process can be applied to a variety of commercial operations where AI and machine synchronization play a vital role. For example, autonomous warehouse delivery robots can be trained – through their own sensors and AI computer – to communicate with other robots within the fleet via the IoRT platform’s ability to crunch data over the cloud. This could be used to pinpoint any high-risk collision hotspots within the robots’ delivery journey and adjust the route which is then sent as a software update within the fleet to avoid future collisions.

Yet these machine-to-machine (M2M) interactions must happen very quickly in order to work, as low latency is key to unlocking the very high volumes of data required with minimal delay. Getting a software ‘task’ update back to a robot fleet to respond to a ‘learning’ or improve on operational protocol demands a reliable and fast solution. Distributed IT infrastructure provides the link between the robot fleet and a point of connectivity to a large processing bandwidth in the cloud. Vendor-neutral interconnection solutions, such as those on Platform Equinix, bridge the gap between the cloud compute power needed for core AI models and the reach required to privately connect to digital ecosystems for low-latency and secure data exchange.

Many firms were already planning a major technological transformation to accommodate this before the arrival of COVID-19. Now as lockdown protocols decrease, organizations are using this as an opportunity to explore ways to reshape their business. Many look to streamline and future proof their workflows by incorporating more automated structures which are less reliant on a human workforce and thus more resilient in the event of a second wave of the outbreak.

Yet as businesses go through the process of evaluation, they should consider whether their automated technology has the ability to bi-directionally learn. Learning robots are more adaptable to AI algorithms in the cloud which ultimately allows businesses increased agility and flexibility to respond to unforeseen events and to better manage workflows as they navigate their way through the challenging months and years ahead.

Mark Anderson is Senior Director of Global Solutions Enablement EMEA, at global interconnection and data centre company, Equinix.

Keep up with the ever-evolving AI landscape
Unlock exclusive AI content by subscribing to our newsletter!!

You May Also Like