There are certainly roles that robots can perform more effectively than humans, including repetitive, monotonous tasks, handling heavy loads or tasks in environments that are dangerous for humans. At the moment, robots only do what they have been programmed to do. The ability to think and make situational decisions in real time is reserved for humans. Artificial intelligence still has a long way to go before it can come close to replicating this.

The emergence of collaborative robots, known as cobots, which are designed to interact with humans in a shared space or work safely in close proximity, is an important step towards human-robot collaboration. Cobots stand in contrast to traditional industrial robots, which work autonomously and whose safety is ensured by isolating them from human contact. Meanwhile, manufacturing companies are beginning to rethink processes in order to use humans and robots together more efficiently.

Cobots need much tighter control for real-time implementation of complex decisions in co-working environments. Artificial intelligence can help here.

Quick decision at the Edge

However, the complex decisions have to be made by the robot as an edge device. As an edge component, it can achieve the required speed and latency to process the increasing amount of data from IoT sensors.

However, it is important to clarify what 'edge' means. An edge device is a component that analyzes multiple data streams and makes decisions based on this data. It can therefore be a server blade in which several information streams converge, a simple gateway or a robot.

If you delve deeper into the architecture of these robots, these products are based on multicore processors. In a number of cases, the processing functions are very similar to those of a smartphone. One of the main differences with robots is that they have to perform their functions in a very predictable, short time. When robots are in close proximity to humans and/or very expensive materials, the robot must react to an event within 100 to 200 µs. This is relatively easy to realize if this is the only task for the processor.

Real-time systems in industry are usually based on a single-core processor (SCP). For the development of such systems, the industry has introduced a process based on the assumption of constant maximum runtime (worst case execution time, WCET). This states that the measured maximum runtime of a software task - should it be executed on a single core alone - remains the same if this task is executed together with other tasks.

Requirements for mixed-critical systems

System architecture for the secure deployment of mixed-criticality applications in a distributed mission-critical edge context.

© Lynx

As the industry moves to so-called 'mixed criticality' systems - those that run real-time tasks in conjunction with less time-critical applications - the assumption of a constant WCET for single-core chips is correct. For current multi-core processors, it does not apply due to interference between cores when accessing shared resources. Without going into deep hardware details, current multicore processors exhibit non-deterministic memory access delays when competing for shared resources. Disruptions can be caused by both hardware and software level contention. This paper focuses on the effects of hardware interference and the so-called interference channels that manifest themselves in various aspects of multicore architectures.

The problem of multicore interference channels typically results from a mixture of:

• Time separation - A temporal separation supported by a scheduling algorithm that can activate cores over non-overlapping time intervals, or by bandwidth partitioning for accessing shared memory resources.
• Space separation - Spatial partitioning of shared resources on the cores to avoid or reduce conflicts over shared resources.

Another requirement for these mixed-critical systems is virtualization. Hardware virtualization involves creating a software environment that mirrors the underlying hardware capabilities so that operating systems themselves can run in a hardware environment other than the original one. The software programs that provide such capabilities are hypervisors. A hypervisor abstracts the capabilities of the hardware and allows multiple, potentially heterogeneous operating system instances to run on a single hardware platform.

There are two main types of hypervisors:

• Type 1 (bare metal) - The 'Type 1' hypervisor runs directly on the hardware and has complete control over the platform.
• Type 2 (Hosted) - The 'Type 2' hypervisor runs on a host operating system and is dependent on the host operating system to control the hardware.

Outsourcing security functions

For a robot system to be highly reliable, care must be taken to isolate different systems from each other so that an application cannot inadvertently or deliberately cause other parts of the system to fail. Effectively, an isolated context is created for each operating system that is currently running. The increasing number of cyber-attacks has shown that if a worm can gain access to a traditional operating system, it will also gain access to the 'crown jewels' of the system. Lynx's approach is therefore to outsource the security policies and hardware authorizations to a separate virtual machine.

An example: A hypervisor responsibly assigns which hardware a particular virtual machine can access. The hardware that provides Internet connectivity can therefore not have access to the network interface that provides access to mission-critical machines. Once these configurations are defined, they must be unchangeable. There must be no root access in the system to change things. Similarly, other system secrets are partitioned so that only
partitioned so that only trusted applications can access them. Some users create bare-metal applications to perform these functions, using memory that is inaccessible to other applications.

More senses for robots

Ian Ferguson is Vice President Strategic Alliances and Marketing at Lynx Software Technologies.

© Lynx

One way in which robots will continue to develop is by integrating sensors into industrial robots. Today, robots mostly only use vision systems. Gesture recognition is already being used in some cases. Soon, however, robots will also use other senses. For example, robots will be able to respond to voice commands and recognize noises that indicate malfunctions in the production plant. It is also conceivable that they will be able to confirm by touch whether a particular product is smooth enough. With the help of special sensor technology, they could 'taste test' the mixtures of certain compounds. Artificial intelligence will help with this: In the future, robots' capabilities will improve the more they are integrated into a factory environment. They will learn to recognize scenarios and react to them.

Many planned next-generation cobot installations will have a connection to a public or private cloud. Systems that only connect to local computing resources increase security on the one hand, but make the self-learning processes of AI more difficult. The right balance is a combination of local learning in conjunction with leveraging the scale performance achieved by aggregated learning accumulated in the cloud.