Our research focuses on autonomous agents and intelligent robots that plan and act under uncertainty to accomplish complex tasks. We are particularly interested in aspects relating to reliability and generalizability of methods for computing the behavior of autonomous agents, going from theoretical formulations to executable systems. Our methods draw upon formal foundations of mathematical logic, probability theory, machine learning, and well-founded notions of state and action abstractions.

Research

Planning and Reasoning Under Uncertainty

A real robot never has perfect sensors or actuators. Instead, an intelligent robot needs to be able to solve the tasks assigned to it while handling uncertainty about the environment as well as about the effects of its own actions. This is a challenging computational problem, but also one that humans solve on a routine basis (we don't have perfect sensors or actuators either!).

We are developing new methods for efficiently expressing and solving problems where the agent has limited, incomplete information about the quantities and identities of the objects that it may encounter.

POMDP Example
Open-universe tiger POMDP with an unknown number of moving tigers.

Generalized Planning

"Be wise, generalize!"

Planning is well known to be a hard problem. We are developing methods for acquiring useful knowledge while computing plans for small problem instances. This knowledge is then used to aid planning in larger, more difficult problems.

Often, our approaches can extract algorithmic, generalized plans that solve efficiently large classes of similar problems as well as problems with uncertainty in the quantities of objects that the agent needs to work with. The generalized plans we compute are easier to understand and are generated with proofs of correctness.

PR2 does the laundry using a generalized planner with our integrated task and motion planning system.

Synthesis and Analysis of Abstractions for Autonomy

It would be impractical to reason about multi-step tasks such as setting a table for dinner at the lowest level of modeling (e.g.,individual joint movements). In order to build autonomous systems that help in complex, multi-step tasks (where we may actually need help), we draw upon the best example of intelligent systems we know: humans. As humans we rarely think about where to place our knees and elbows. Instead, we reason about achieving abstracted regions of configurations (e.g. reach the 5fth floor).

Automatically coming up with the right abstraction to solve a problem efficiently is a notoriously challenging problem. We are developing new methods for utilizing abstractions in sequential decision making (SDM), for evaluating the effect of abstractions on models for SDM, as well as to search for abstractions that would aid in solving a given SDM problem.

Maze Abstraction into Rooms
Abstraction of maze into rooms

Integrated Task and Motion Planning

In order to solve tasks such as doing the laundry, a robot needs to compute high-level strategy (should I use the basket?) as well as the joint movements that it should execute. Unfortunately, approaches for making high-level decisions rely on "task planning" abstractions that are lossy and can produce strategies that have no feasible motion plans.

We are developing new methods for dynamically refining the task-planning abstraction to produce combined task and motion plans that are executable.

PR2 sets the table for dinner using our integrated task and motion planning system.

PR2 (in simulation) grasps an object in clutter using the same integrated task and motion planning system.

Autonomous Agents That Are Easy to Understand and Safe to Work With

AI systems have the potential to improve our society in many walks of life. However, today's AI systems require highly trained experts for their customization, configuration, and repair. This not only makes it difficult to realize the potential benefits of AI in society, but also creates large uncertainties in the future of employment for millions in the workforce. On the other hand, perfectly transparent AI systems would present grave risks wherever information needs to be protected, as they may inadvertently reveal sensitive information (thus compromising user privacy or protected data) and may be susceptible to disruptive adversaries.

To address these issues, we are developing new ways for AI systems to explain their behavior in a manner aligned with the proficiency of the users. We are also developing methods for computing agent behavior that reduces the amount of information yielded to adversaries while maximizing its understandability by allies. These methods would facilitate spontaneous, productive teamwork between AI systems and people who may be experts in fields other than AI.

Assistive Planning
Examples of explanation and behavior computation.

Roblocks: An Educational System for AI Planning and Reasoning

The objective of this project is to introduce AI planning concepts using mobile manipulator robots. It uses a visual programming interface to make these concepts easier to grasp. Users can get the robot to accomplish desired tasks by dynamically populating puzzle shaped blocks encoding the robot's possible actions. This allows users to carry out navigation, planning and manipulation by connecting blocks instead of writing code. AI planning techniques are used to prune down the vast number of actions possible to suggest only the truly feasible and relevant actions.

Roblocks System
The Roblocks system in action.
    DARPA logo         NSF logo    

People



Siddharth Srivastava

Siddharth Srivastava

Assistant Professor
Director, AAIR Lab




Rushang Karia

Rushang Karia

PhD Student

Naman P Shah

Naman P Shah

PhD Student

Pulkit Verma

Pulkit Verma

PhD Student




Chirav Dave

Chirav Dave

MS Student

Kislay Kumar

Kislay Kumar

MS Student

. Daniel Molina

Daniel Molina

MS Student




Rashmeet K Nayyar

Rashmeet K Nayyar

MS Student

Julia Nakhleh

Julia Nakhleh

BS Student

Alumni


Midhun P M MCS Dec 2017 - Dec 2018
Ryan Christensen BS Aug 2017 - May 2018
Perry Wang BS Aug 2017 - May 2018

AAIR Lab Group - Fall 2018

Publications


Show by area:

  • Tractability of Generalized Planning.
    Siddharth Srivastava, Shlomo Zilberstein, Abhishek Gupta, Pieter Abbeel, Stuart Russell.
    In Proceedings of AAAI, 2015.
    Partial Observability Learning State/Action Abstractions Generalized Planning Plan Generalization and Transfer
  • First-Order Open-Universe POMDPs.
    Siddharth Srivastava, Stuart Russell, Paul Ruan, Xiang Cheng.
    In Proceedings of the Conference on Uncertainty in AI (UAI), 2014.
    Partial Observability Probabilistic Inference
  • Qualitative Numeric Planning.
    Siddharth Srivastava, Shlomo Zilberstein, Neil Immerman, Hector Geffner.
    In Proceedings of the Twenty Fifth Conference on AI (AAAI), 2011.
    Partial Observability Generalized Planning State/Action Abstractions Plan Generalization and Transfer
  • Computing Applicability Conditions for Plans with Loops. [Best Paper Award] (TechReport with more results and detailed proofs)
    Siddharth Srivastava, Neil Immerman, Shlomo Zilberstein.
    In Proceedings of the Twentieth International Conference on Automated Planning and Scheduling (ICAPS), 2010.
    Generalized Planning State/Action Abstractions Plan Generalization and Transfer
  • Finding Plans with Branches, Loops and Preconditions.
    Siddharth Srivastava, Neil Immerman, Shlomo Zilberstein.
    ICAPS 2009 Workshop on Verification and Validation of Planning and Scheduling Systems, 2009. [slides]
    Generalized Planning State/Action Abstractions Plan Generalization and Transfer
  • Challenges in Finding Generalized Plans.
    Siddharth Srivastava, Neil Immerman, Shlomo Zilberstein.
    ICAPS 2009 Workshop on Generalized Planning: Macros, Loops, Domain Control, 2009. [slides]
    Generalized Planning State/Action Abstractions Plan Generalization and Transfer
  • Using Abstraction for Generalized Planning.
    Siddharth Srivastava, Neil Immerman, Shlomo Zilberstein.
    International Symposium on AI and Mathematics (ISAIM), 2008.
    State/Action Abstractions Plan Generalization and Transfer Generalized Planning

Contact Us



BYENG Building

Google Scholar  Twitter    aair.lab@asu.edu

Brickyard Engineering #507, 699 S. Mill Ave., Tempe, AZ - 85281