OptaPlanner: The open-source tool that can automate schedules

21 Jun 2022

Image: © tatomm/Stock.adobe.com

After Geoffrey de Smet learned about the concept of rule engine algorithms, he used his knowledge to optimise one of the most common administrative tasks.

Click to read more stories from Automation Week.

One of the most important elements of automation is optimisation. Workers across multiple industries want to use it to cut down on monotonous, administrative tasks that can be carried out by intelligent algorithms – leaving humans to do the tasks that require more thought or creativity.

For those who don’t work directly with algorithms and code, they need some sort of automation tool to achieve this. That is where people like Red Hat’s Geoffrey De Smet come in.

De Smet is the lead and creator of OptaPlanner, an open-source AI constraint solver that can be used to solve planning problems and automate schedules such as vehicle routing, employee rostering and maintenance scheduling.

“In 2006, I was working in a research group that focused on metaheuristics and other AI algorithms,” he told SiliconRepublic.com.

“A presentation at a local conference taught me about rule engine algorithms. I was inspired to combine them. A long vacation later, OptaPlanner was born.”

Creating OptaPlanner

De Smet worked on OptaPlanner for many years in his spare time. He would often join academic operations research challenges to see if he could solve them with OptaPlanner.

“After all, someone had to help Santa Claus find the shortest path to visit all kids on this planet,” he said. “These competitions regularly exposed me to better algorithms and implementation techniques, which I quickly assimilated into OptaPlanner.

“For example, in a competition around 2012, a team used the late acceptance algorithm to beat OptaPlanner’s results. This metaheuristics algorithm, invented by Yuri Bykov, is on average typically better than Tabu Search and Simulated Annealing. So I implemented it for OptaPlanner too.”

De Smet joined Red Hat in 2010 and by 2013, the open-source software company had started productising OptaPlanner and offering enterprise support for it. “My hobby became my full job,” he said.

The use cases of this automation are now far-reaching. OptaPlanner is able to reduce the driving time of a fleet of vehicles by deciding which vehicle goes to where and in which order.

In employee scheduling for shift workers such as nurses, doctors and security guards, the algorithm assigns every shift to an employee taking into account skills, affinity, availability and other constraints.

“Other major use cases include maintenance scheduling, school timetabling, order-picking routing, job shop scheduling, and court hearing scheduling,” De Smet said.

The future of automation

While automation will be able to optimise a lot of work across a wide range of industries, De Smet said businesses will need to be able to adjust frequently and quickly in order for automation to be truly effective.

“For example, a machine learning model trained on last year’s flight data might not be relevant today, now that tourism is increasing again,” he said.

“Another big trend I see is the need to clearly measure the return on investment (ROI) of any AI technology implementation. The time of hand waving is over. At the same time, the ROI of many AI projects is huge, but so is the leap to take them. Often, the returns can’t be materialised in small, incremental steps, only at the end when it fully works – or it doesn’t.”

With this potential all-or-nothing result, De Smet said the experience of pay-as-you-go for AI development is something that needs to change.

He also said that one of the biggest challenges in the AI industry as a whole is to convince users that this area of tech is more than just machine learning and that it’s vital that the right tool is used for the right job.

“Machine learning, and deep learning neural nets in particular, are great for pattern recognition: image recognition, voice recognition and similar – things that humans are good at,” he said.

“Machine learning is consistently inferior in planning and scheduling. Use metaheuristics and mathematical optimisation algorithms for such use cases. Hammering a screw leads to suboptimal results.”

10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.

Jenny Darmody is the editor of Silicon Republic