Alberto Camacho

Research Interests

Decision-Theoretic Planning

We investigate effective methods to generate policies in stochastic environments, typically modeled as MDPs.

  • Automata-Based Reward Shaping for MDPs with LTL-f goals. [SoCS-17,RLDM-17]
  • Probabilistic Planning for reachability goals. [ICAPS-16]
  • Related Papers
  • [SoCS-17] Non-Markovian Rewards Expressed in LTL: Guiding Search via Reward Shaping.
  • [RLDM-17] Decision-Making with Non-Markovian Rewards: From LTL to automata-based reward shaping.
  • [ICAPS-16] From FOND to Robust Probabilistic Planning: Computing compact policies that bypass avoidable deadends.
  • Model Learning

    We investigate effective methods to learn interpretable models from data.

  • Learning Interpretable Models in Linear Temporal Logic. [ICAPS-19]
  • Related Papers
  • [ICAPS-19] Learning Interpretable Models in Linear Temporal Logic.
  • Program Synthesis for LTL Specifications

    Everybody should be able to tell their machines what to do. Software synthesis is the problem of constructing software (i.e. a program) from user intent. Our work investigates automatic construction of reactive programs that run in perpetuity, and also programs that terminate. We use Linear Temporal Logic (LTL) as specification language.

    LTL Synthesis as a Service
  • SynKit is a web-service to perform rapid synthesis on the cloud. [IJCAI-18demo]
  • LTL Realizability and Synthesis
  • We present novel duality results between LTL realizability and reachability games. [IJCAI-18]
  • We present the first algorithmic approach to LTL realizability and synthesis via automated AI planning. [IJCAI-18]
  • Finite LTL Synthesis
  • We motivate the need for environment assumptions in Finite LTL Synthesis. [KR-18]
  • We study synthesis of high-quality strategies. [KR-18]
  • We introduce certificates of unrealizability for Finite LTL Synthesis. [ICAPS-18]
  • We present the first algorithmic approach to LTL realizability and synthesis via automated AI planning. [ICAPS-18]
  • LTL Synthesis and AI Planning
  • We establish a clear correspondence between LTL Synthesis and AI Planning. [ICAPS-19]
  • Preliminary results on the correspondence between LTL Synthesis and AI Planning. [Canadian AI'18]

  • Related Papers
  • [ICAPS-19] Towards a Unified View of AI Planning and Reactive Synthesis.
  • [KR-18] Finite LTL Synthesis with Environment Assumptions and Quality Measures.
  • [IJCAI-18] LTL Synthesis via Safety and Reachability Games.
  • [IJCAI-18demo] SynKit: LTL Synthesis as a Service.
  • [ICAPS-18] Finite LTL Synthesis as Planning.
  • [Canadian AI'18] Synthesizing Controllers: On the Correspondence Between LTL Synthesis and Non-deterministic Planning.
  • [Genplan-17] Bridging the Gap Between LTL Synthesis and Automated Planning.
  • Non-deterministic Planning with Temporally Extended Goals

    In Fully Observable Non-deterministic (FOND) Planning, the effects of the actions that that an agent may perform are non-deterministic. We study Fully Observable Non-deterministic Planning with Temporally Extended Goals (AAAI-17 paper). Our techniques compute finite, and infinite plans whose execution satisfy a given property that is temporally extended, and that are robust to the non-determinism in the effect of the actions.

    In our RLDM-17 and SoCS-17 papers we introduce effective techniques to guide search in MDPs with non-Markovian rewards.


    Related Papers
  • [AAAI-17] Non-Deterministic Planning with Temporally Extended Goals: LTL over Finite and Infinite Traces.
  • [SoCS-17] Non-Markovian Rewards Expressed in LTL: Guiding Search via Reward Shaping.
  • [RLDM-17] Decision-Making with Non-Markovian Rewards: From LTL to automata-based reward shaping.
  • LTL FOND to FOND Repository

    Probabilistic Planning

    Our ICAPS-16 paper introduces ProbPRP, the state-of-the art in probabilistic planning. ProbPRP computes policies whose execution attempt to maximize the probability to achieve a prescribed goal.


    Related Paper
  • [ICAPS-16] From FOND to Robust Probabilistic Planning: Computing compact policies that bypass avoidable deadends.
  • ProbPRP Repository

    Software

    SynKit, a web-service and API for rapid synthesis of LTL and Finite LTL specifications.
    SynKit
    ProbPRP, a probabilistic planner for HIGHPROB with state-of-the-art performance.
    ProbPRP Repository
    LTL FOND to FOND, compilations of FOND planning problems with temporally extended goals expressed in LTL into standard FOND planning with final-state goals.
    LTL FOND to FOND Repository

    Other Posters and Talks

    From Logistics to Drones: Customized Controllers for Autonomous Systems,
    In Amazon's Research Day, University of Toronto, 2015.
    Poster
    Synthesizing Finite and Infinite Plans with Temporally Extended Goals,
    In Research in Action. University of Toronto, 2016.
    Poster Abstracts
    Computing High-Quality Solutions to Probabilistic Planning Problems,
    In Research in Action. University of Toronto, 2015.
    Poster
    Computing Compact Policies for FOND Planning Problems,
    In EASSS 2013, London.
    Poster Abstracts
    Artificial Intelligence: Agents on Earth and Space.
    In ESA Summer Alumni Meeting 2013, Madrid.

    Community Service

    Program Committee:
    • AAAI-19
    • AAAI-19 Student Abstracts
    • ICAPS 2017 Workshop on Generalized Planning (GenPlan-17)

    Reviewer:
    • ICAPS (2018)

    Subreviewer:
    • AAAI (2015, 2016, 2017)
    • IJCAI (2015, 2016, 2018)
    • ICAPS (2016, 2017)
    • KnowProS (2017)