Workshop on A Reality Check for Planning and Scheduling Under Uncertainty
Research on planning and scheduling under uncertainty is one of most active areas in AI today. Many applications motivate the need to cope with uncertainty due to effectors, sensors, environment, adversaries, allies, and even the imperfect planning agent. However, the benefit of techniques falling within this category is highly contentious for many (provocative) reasons:
- The scalability of probabilistic/decision theoretic planners is a far cry
from that of deterministic planners.
- Re-planners dominate conditional (pre)planners in the IPC.
- Several domains involving uncertainty are not probabilistically interesting, and hence, deterministic planners suffice for such domains.
- Acquiring models (e.g., probability distributions) from humans is difficult
and often times subjective, whereas learning models can sometimes be more successful with hard-to-come-by, but good data.
- Much work concentrates on optimal (or approximately optimal) solutions, despite the limited success of such techniques in even deterministic settings.
- Models of the uncertainty in real-world scheduling domains are often so
poor, or the variability between instances so great, that complex anticipatory scheduling approaches render schedules brittle and suboptimal with hindsight.
- The research community has studied abstracted benchmarks and produced dedicated algorithms that fail to impact real-world scheduling instances. Practitioners favor straightforward deterministic scheduling techniques combined with online schedule refinement and repair.
This workshop invites technical and opinion papers that help conduct a reality check on planning and scheduling under uncertainty to see what advances are being made, as well as what outstanding challenges remain.
Topics include, but are not limited to:
- Responses to the provocations above: for or against.
- Models and languages for planning and scheduling under uncertainty.
- Knowledge acquisition for models of uncertainty.
- Replanning versus Conditional (Pre)Planning.
- Algorithm selection based on domain analysis.
- Scaling planning and scheduling under uncertainty.
- Applications of planning and scheduling under uncertainty.
- Execution and execution monitoring in uncertain environments.
- Adversarial or multi-agent techniques.
Organizers
- Daniel Bryce, Utah State, bryce at ai.sri.com
- Mausam, University of Washington, mausam at cs.washington.edu
- Sungwook Yoon, Arizona State University, Sungwook.Yoon at asu.edu
