More about the machine learning addition to the Autodesk tool.
Amongst a number of new features being announced for Shotgun, Autodesk has indicated it is testing a generative scheduling based on the acquisition of technology known as Consilium.
The idea is to enable more accurate bidding, scheduling and planning with the tool using a form of machine learning to automatically produce a schedule. befores & afters asked Autodesk senior software architect Phil Peterson to explain further.
b&a: Can you explain what Consilium’s tech is, and how it is going to form part of Shotgun? What training data etc is required to make this a machine learning tool?
Phil Peterson (senior software architect, Autodesk): This technology allows you to take a production plan—the tasks, their resource requirements, dependencies and time constraints, and rapidly generate complete, actionable schedules optimized for resource utilization. And, the workflow allows you to create and compare multiple scenarios to explore different trade-offs in your planning without disrupting your live production schedule until you are ready to commit.
The technology is currently being integrated as part of the Shotgun ecosystem so that a production scheduler can start with their tasks and resources in Shotgun, then create and explore scenarios and schedules before updating their Shotgun production schedule with their new, chosen and optimized result.
No training data is required for this to work. The technology applies a form of machine learning built upon a bespoke evolutionary computation engine that learns how to make the decisions that a production manager would need to make in order to meet the constraints of a plan and optimize it. It does this solely based on the given project description, recognizing that each production is unique.
b&a: What does this mean, day to day, for VFX producers and crew in their use of Shotgun?
Phil Peterson: For producers and production management, it can greatly accelerate the time needed to create and maintain a good schedule, and very likely one that is more optimal than would be practical to do by hand. Because of this speed, it means that they are able to consider multiple variations and what-if scenarios to make better high-level choices and be in a position to both be more predictive and adapt to inevitable change.
b&a: How, in general, have clients been using it?
Phil Peterson: Clients have been using this technology to initially plan for a production based on their tasks in Shotgun. Especially to determine and optimize the number of crew needed including considering tradeoffs in different department sizes and timing. But, they have also been using it to re-schedule remaining work when circumstances change, as they inevitably do. In some cases, what was taking weeks of manual editing can be accomplished in a few hours, or even minutes of exploration and frequently with a more optimal result than their efforts by-hand. Clients such as LAIKA, do this with thousands of tasks and hundreds of inventory and delivery constraints for upstream and downstream production requirements.
You can find out more about this new feature in Shotgun at Autodesk’s website, along with other Shotgun features such as Story-in-Context that provides on-set and editorial data direct on the editorial timeline, and the move to enable open standards-based asset management across the life-cycle of an entire production.