Implementing Advanced Vision Systems for Safer Robotics

In most cases, robots improve workplace safety. Manufacturers typically automate some of the most dangerous tasks, effectively eliminating hazards by placing distance between them and workers. Despite that advantage, industrial robots carry some safety risks of their own.

Robots’ inability to adapt to unexpected conditions or to see incoming obstacles can cause collisions or other hazardous situations. Manufacturers must address these risks to use robotics to its full potential. Advanced vision systems can provide a means to that end in several ways.

Mobile Robot Navigation

The most straightforward safety application of machine vision in robotics is guiding mobile robots. Mobile robots are one of the fastest-growing industrial automation segments, with experts expecting companies to deploy thousands by 2027. As this equipment becomes more common, run-ins with human workers become a bigger concern, but machine vision enables safer navigation.

Many automated guided vehicles (AGVs) and similar machines follow set paths or use wireless beacons to navigate. While efficient, this approach cannot account for workers crossing robots’ paths and creating collision risks. Adding machine vision into the mix ensures that AGVs can recognize people and other obstacles to adjust their route in real-time and avoid impacts.

Modern machine vision algorithms can classify different objects in fractions of a second. This speed and accuracy lets AGVs take different approaches depending on the situation. For example, they may go around a stationary obstacle but stop for people until they pass.

Stationary Spacial Awareness

Fixed robotic systems are inherently safer than mobile alternatives because of their limited movement. However, they can still move too fast or far and hit nearby workers. Machine vision provides a solution through real-time hazard detection.

Cameras around a stationary robot can monitor workers who may approach it. Machine vision algorithms then analyze this data to detect when humans are too close to operate safely, halting or adjusting the robot’s motion until the person moves a safe distance away.

Proximity sensors enable similar protections, but machine vision is more reliable because it distinguishes between people and other nearby objects. Vision is also preferable to conventional physical barriers, providing similar protection without occupying more space.

Error Prevention

Organizations can also use vision systems to improve safety by making robots’ work more accurate. Many workplace hazards arise from error, and while automation typically reduces these mistakes, conditions must remain predictable. Machine vision enables adaptability to ensure accuracy and safety despite changing circumstances.

Consider welding automation, for example. Failing to account for surface flaws, improper alignment or unusual shapes can result in more sparks, fumes and similar hazards. These byproducts may fly up to 35 feet away, endangering workers even though the robot handles the job’s most hazardous part.

3D vision can account for these flaws or unexpected changes that conventional 2D systems may miss. The robot welder can then adapt its approach to complete the weld with minimal sparking or other hazards. This increased accuracy will also decrease material-related risks in downstream processing.

How to Implement Vision Systems for Improved Safety

As beneficial as these use cases are, they all rely on proper implementation to achieve optimal results. Like robotics, advanced vision systems are only as effective as their users’ ability to apply best practices.

Choose the Right Machine Learning Model

First, it is important to recognize that the ideal vision system depends on the task. This optimization begins with choosing the suitable underlying machine learning model.

Logistic regression algorithms excel at either-or tasks and are relatively easy to train, making them conducive to binary vision applications. These may include determining if a part is properly aligned before welding or stopping a robot if an obstacle is nearby. However, more nuanced tasks, like guiding an AGV through a warehouse, require more complex algorithms.

Random forests and deep learning models are more accurate and versatile but require more training and computing power. Manufacturers can determine the right kind of model by recognizing the specific purpose of their desired vision system. They can then refine their choices by comparing applicable models to their budget, processing power and artificial intelligence.

Ensure Appropriate Lighting

Next, organizations must ensure their vision systems have an operating environment that promotes accuracy and reliability. That means providing sufficient lighting, as machine vision, like humans, can recognize objects more accurately when well-lit.

Optimal lighting varies between use cases. Directional lighting is ideal when maximizing contrast at predictable angles but can create accuracy-affecting hotspots. Manufacturers must consider different lighting angles, sources and frequencies to maximize contrast where desired and minimize it elsewhere.

Free-roaming robots like AGVs don’t need extreme lighting angles. Ensuring consistent overhead lighting throughout the facility is enough. Stationary robots can use shielding and on-device lights for more case-specific lighting needs.

Start Small

Machine vision can be challenging, even when organizations give enough thought to lighting and model selection. Consequently, it’s also best to test these systems in small use cases before expanding them.

Start with an application where errors will not result in significant hazards. Ideally, this use case should be non-mission-critical and carry relatively low ongoing costs. Pay attention to where the vision system excels, what causes performance issues and what changes fix those problems.

Once this initial rollout meets satisfactory benchmarks, manufacturers can expand machine vision to more applications. This restrained approach to adoption may take more time to deliver significant results. However, applying vision systems to more mission-critical processes ensures a smoother, more cost-effective rollout.

Optimize Over Time

Similarly, ongoing testing and optimization are necessary to ensure machine vision systems are as safe as possible. Supervised machine learning models and reinforcement algorithms increase accuracy over time as they encounter more data, but this process is often not automatic. Organizations must test and refine the underlying model to address problems as they arise and hone their reliability.

Vision systems that excel in training scenarios will likely still encounter issues eventually. No system is perfect, just as no employee performs perfectly every day. Consequently, manufacturers must always monitor, benchmark and adjust machine vision solutions so they can reach their full potential.

Advanced Vision Systems Make Robots Safer

Robots could revolutionize workplace safety. For that to happen, though, the organizations employing this technology must equip it with tools to help it recognize and respond to hazards. That means capitalizing on machine vision.

As vision technology improves, its potential for robot safety will grow. Staying abreast of these advances will allow manufacturers to become as efficient and safe as possible.