Multi-Agent Pathfinding: Models, Solvers, and Systems

Roman Barták • Charles University, Czech Republic

Philipp Obermeier • Inria Rennes, France

Torsten Schaub • University of Potsdam, Germany

Tran Cao Son • New Mexico State University, USA

Roni Stern • Ben Gurion University, Israel

The topic of multi-agent pathfinding (MAFP) has become very important with the rise of multi-robot systems in areas such as warehousing and autonomous vehicles, where collision free paths need to be planned. The aim of the tutorial is to provide a survey of MAPF problems and techniques for solving them. Practical motivation and applications will also be discussed together with demos of benchmarking systems (MAPF Scenario and ASPRILO). The target audience is anyone interested in planning for multiple agents. The attendees will walk away with information about the state-of-the-art in MAPF.

Planning and Scheduling Approaches for Urban Traffic Control

Scott Sanner • University of Toronto, Canada

Stephen F. Smith • Carnegie Mellon University, USA

Mauro Vallati • University of Huddersfield, United Kingdom

The goal of the tutorial is to give an overview of the Planning and Scheduling techniques that have been designed to support Urban Traffic Control. In a nutshell, the tutorial provides an introduction to Urban Traffic Control and to Planning and Scheduling, and describes three different techniques that have been designed for dealing with issues of Urban Traffic Control. The tutorial is of an introductory nature – only the basic knowledge of planning and scheduling is required – and thus will be accessible to a wide audience.

Temporal Reasoning

Nikhil Bhargava • Brian Williams • Massachusetts Institute of Technology, USA

In the tutorial, I plan to cover the basics of reasoning in the context of temporal networks with a particular focus on temporal networks with uncertainty. The high-level goal of the tutorial is to introduce temporal reasoning from the lens of someone who is considering creating or modifying a planner that is capable of reasoning about temporal uncertainty. The plan is to start with a high-level summary of several existing types of temporal networks with a particular focus on the tradeoff between the expressivity of the network and the complexity of finding a valid schedule. The three types of networks that I will focus on in particular are Simple Temporal Networks, Simple Temporal Networks with Uncertainty, and Probabilistic Simple Temporal Networks. In addition to standard best-practice algorithms for finding schedules, I will also explore incremental consistency checking, i.e. what happens when individual constraints are added and removed.

AI Planning for Robotics with ROSPlan

Michael Cashmore • Daniele Magazzeni • King's College London, United Kingdom

This tutorial will provide an introduction to AI Planning & Scheduling (AIP&S) for robotics, the essentials of the Robot Operating System (ROS) and ROSPlan. This tutorial will also provide an overview of recent advances in integrating AIP&S and robotics. The main goal is to illustrate the integration of AIP&S and robotics, describe how it is beneficial to both communities, as well as to highlight the main challenges and open issues. The tutorial will cover different aspects of this integration, including modelling domains for robotics, strategies for how to use planning effectively, linking plans to robot actions and sensing, and plan execution in an uncertain environment. Through this tutorial we plan to motivate and encourage the use of AIP&S for robotics applications, and to equip new researchers with the knowledge required for applying new approaches in those applications through existing tools, such as ROSPlan. We believe that such a tutorial will help in fostering the development of AIP&S techniques for robotics. Moreover, the tutorial aims to bring challenges and expectations related to planning for autonomous robots (deliberative, reactive, continuous planning and execution etc.) to the AIP&S community.

From Teaching the PDDL Novice to Empowering the Planning Solution Integrator

Jan Dolejsi • Derek Long • Maria Fox • Schlumberger, UK

Christian Muise • IBM Research AI, Cambridge, USA

This tutorial will show a suite of tools for PDDL educators, modelers, solution developers and planning algorithm researchers/implementers. The tutorial will show how to start with the online editor.planning.domains and gradually transition to an offline integrated developer environment to build more powerful planning solutions, or diagnose planner performance.

Integrated Tasks and Motion Planning

Malik Ghallab • Felix Ingrand • Rachid Alami • Thierry Simeon • LAAS-CNRS, Toulouse, France

The need to integrate motion, manipulation and task planning has been recognized since the beginning of planning research in robotics. Because of the frequent coupling between the tasks and the precise motions for achieving them, this integration is demanded in most interesting robotics applications where any planning is needed. A decomposition which leaves motion issues at the execution level is seldom satisfactory. There is an increasing number of publications addressing Integrated Tasks and Motion Planning (ITAMP) problems, with dedicated events in many Robotics and AI conferences. The topic is becoming rich, but it is also blurred by a wide diversity of the technical innovations in various directions, some very promising and opening important perspectives, other more incremental. This tutorial aims at giving to the ICAPS community (assumed familiar with task planning) the needed background about motion and manipulation planning as well as a survey of the main approaches addressing ITAMP. It should be noted that the ITAMP integration problem is of interest to planning beyond the robotics domain, since other planning areas may require "metric fluents", usually dealt with in ad hoc manners.

Goal Recognition Design

Sarah Keren • Harvard University, USA

William Yeoh • Washington University in St. Louis, USA

Goal recognition is the task of understanding the goal of an acting agent by the online observation of its behavior. Goal recognition design (GRD) facilitates goal recognition by the analysis and redesign of goal recognition models. In a nutshell, given a model of a domain and a set of possible goals, a solution to a GRD problem determines: (1) to what extent do actions, performed by an agent, reveal the agent's objective? and (2) what is the best way to modify the model so that the objective of an agent can be detected as early as possible? GRD answers these questions by offering a solution for assessing and minimizing the maximal progress of any agent before recognition is guaranteed. This approach is relevant to any domain for which quickly performing goal recognition is essential and in which the model design can be controlled. Applications include intrusion detection, assisted cognition, computer games, and human-robot collaboration. The tutorial will provide an overview of the GRD problem in both deterministic and stochastic environments, and present the solutions developed so far for its evaluation and optimization. Specifically, we will focus on the relationship between GRD and automated planning, and show how planning is used to both model and solve the GRD problem.

Deep Reinforcement Learning with Applications in Transportation

Zhiwei (Tony) Qin • DiDi Research America, AI Labs Mountain View, California, USA

Jian Tang • DiDi AI Labs, Beijing, China Syracuse University, Syracuse, NY

Jieping Ye • DiDi AI Labs, Beijing, China · University of Michigan, Ann Arbor, MI, USA

This tutorial is targeted to researchers and practitioners with a general machine learning background and are interested in working on applications of deep RL (DRL) in transportation. The goal of this tutorial is to provide the audience with a guided introduction to this exciting area of AI with specially curated application case studies in transportation. The tutorial covers both theory and practice, with more emphasis on the practical aspects of DRL that are pertinent to tackle transportation challenges. After the half-day of lectures, the audience would get an overview of the core DRL methods and their applications, particularly in transportation and ride-sharing domains. They will have a better understanding about the major challenges in transportation and how DRL can help solve those problems. They will also be introduced to several popular open-source DRL development and benchmarking frameworks to get a head-start in experimentation.