homefocus areas › planning and scheduling


planning and scheduling   back to focus areas

AI techniques that allow autonomous agents to act intelligently

Automated planning and scheduling are fundamental AI techniques that allow autonomous agents to act intelligently. To avoid error-prone rules and sub-optimal heuristics, a planning system adopts a declarative approach to problem solving by reasoning over goals, states, and actions. This leads to a fully automated system for synthesizing intelligent behaviors in complex and dynamic environments, often in real time. It brings about a number of significant advantages over traditional rule-based systems, including:


Because the system uses declarative, goal-oriented action synthesis, there will be no need for hand-tailored or hard-coded control rules to exist, dramatically reducing the software (and hardware) development and deployment costs for complex systems. 


A typical system has multiple components, each can fail on its own or as a result of another failure. Thus, the number of possible failure scenarios grows very fast, making it infeasible for humans to hand-craft recovery policies for each and every abnormal condition. Fortunately, using automated planning and re-planning allows us to handle all kinds of exceptions to ensure the robustness of the system under normal or abnormal operating conditions. 


Thanks to the advances in domain-independent planning, a planning system can usually be adapted to multiple domains sharing same or similar constraints. Modern planning languages such as PDDL (Planning Domain Definition Language) are general enough to accommodate a wide variety of domains without issues. This creates more opportunities for reusing existing software to solve problems from new domains, protecting the investment in developing a planning system. 


Many existing systems consider planning and scheduling as two separate steps, which can often result in myopic policies and sub-optimal throughput. Our integrated approach to planning and scheduling performs joint combinatorial optimization to find more globally optimal policies with higher throughput. It also ensures graceful degradation, if some parts of the system cannot function at their 100% capacity. 

Execution monitoring

Because planning is model-based, this allows us to use a similar approach to execution monitoring and plan visualization. The same declarative-programming benefits, including simplicity, robustness, and adaptability, also apply to execution monitoring, as exemplified by the videos shown here. 


A related benefit is the possibility of using a planner to automatically and exhaustively examine all the reachable states of a system to prove the absence of “bad” states, which may include undefined system behavior, racing conditions, unsafe states, and etc. This is another area where machines can easily outperform humans, in terms of the speed and scalability for which the correctness of the system can be thoroughly checked and verified.




Optimizing routes of items through a series of scarce resources.



Rerouting items around errors



Buffering items until resources are available


More videos

Rerouting items around breaks

Modular template

Maximize resource utilization

Mackenzie Venezia


Routing to multiple resources


Tim Curley
Director of Business Development


fact sheets

Intelligent Automation



in the news   view all 

Swinburne Partners With PARC, a Global Leader in Technology Innovation
3 May 2017 | Swinburne University of Technology


blog posts   view all 

A Vision for the Self-Aware Machine
posted 17 December 2013

Sparked by IT megatrends, manufacturers are currently undergoing an operational transformation with increased agility and efficiency as well as fewer operators per machine