• Corpus ID: 221655845

Planning with Learned Object Importance in Large Problem Instances using Graph Neural Networks

@article{Silver2021PlanningWL,
  title={Planning with Learned Object Importance in Large Problem Instances using Graph Neural Networks},
  author={Tom Silver and Rohan Chitnis and Aidan Curtis and Joshua B. Tenenbaum and Tomas Lozano-Perez and Leslie Pack Kaelbling},
  journal={ArXiv},
  year={2021},
  volume={abs/2009.05613}
}
Real-world planning problems often involve hundreds or even thousands of objects, straining the limits of modern planners. In this work, we address this challenge by learning to predict a small set of objects that, taken together, would be sufficient for finding a plan. We propose a graph neural network architecture for predicting object importance in a single pass, thereby incurring little overhead while substantially reducing the number of objects that must be considered by the planner. Our… 

Figures and Tables from this paper

Learning Geometric Reasoning and Control for Long-Horizon Tasks from Visual Input
TLDR
Despite having no explicit representation of the geometry nor pose of the objects in the scene, this network is still able to accomplish geometrically precise manipulation tasks, including handovers and an accurate pointing task where the parameters of early actions are tightly coupled with those of later actions.
SyDeBO: Symbolic-Decision-Embedded Bilevel Optimization for Long-Horizon Manipulation in Dynamic Environments
TLDR
This approach generates highly dynamic manipulation motions by incorporating the full robot and object dynamics and solves a holistically integrated bilevel optimization problem involving costs from both the low-level TO and the high-level search.
PDDLGym: Gym Environments from PDDL Problems
TLDR
PDDLGym is presented, a framework that automatically constructs OpenAI Gym environments from PDDL domains and problems, making the framework particularly well-suited for research in relational reinforcement learning and relational sequential decision-making.
Taskography: Evaluating robot task planning over large 3D scene graphs
TLDR
This work constructs TASKOGRAPHY, the first large-scale robotic task planning benchmark over 3DSGs, and proposes SEEK, a procedure enabling learning-based planners to exploit 3DSG structure, reducing the number of replanning queries required by current best approaches by an order of magnitude.
PG3: Policy-Guided Planning for Generalized Policy Generation
TLDR
This work studies generalized policy search-based methods with a focus on the score function used to guide the search over policies, and demonstrates limitations of two score functions and proposes a new approach that overcomes these limitations.
Efficient and Interpretable Robot Manipulation With Graph Neural Networks
TLDR
In these experiments, a single GNN policy trained using imitation learning (IL) on 20 expert demos can solve blockstacking, rearrangement, and dishwasher loading tasks; once the policy has learned the spatial structure, it can generalize to a larger number of objects, goal configurations, and from simulation to the real world.
Inventing Relational State and Action Abstractions for Effective and Efficient Bilevel Planning
Effective and efficient planning in continuous state and action spaces is fundamentally hard, even when the transition model is deterministic and known. One way to alleviate this challenge is to
Learning Multi-Object Dynamics with Compositional Neural Radiance Fields
TLDR
A method to learn compositional predictive models from image observations based on implicit object encoders, Neural Radiance Fields, and graph neural networks, and a graph neural network dynamics model in the latent space to achieve compositionality for dynamics prediction.
Mission Planning for Underwater Survey with Autonomous Marine Vehicles
TLDR
This study presents an automated planning strategy for a long-term survey mission using a fleet of AMVs consisting of autonomous surface vehicles and autonomous underwater vehicles and demonstrates the efficacy of the proposed methodology.
Solving Disjunctive Temporal Networks with Uncertainty under Restricted Time-Based Controllability using Tree Search and Graph Neural Networks
TLDR
A tree search algorithm to determine whether or not a DTNU is R-TDC and a graph neural network as a heuristic for tree search guidance leads to substantial performance gains on benchmarks of more complex DTNUs, with up to eleven times more problems solved than the baseline tree search.
...
1
2
...

References

SHOWING 1-10 OF 49 REFERENCES
Online Planner Selection with Graph Neural Networks and Adaptive Scheduling
TLDR
A graph neural network (GNN) approach to selecting candidate planners, advantageous over a straightforward alternative, the convolutional neural networks, in that they are invariant to node permutations and that they incorporate node labels for better inference.
Learning How to Ground a Plan - Partial Grounding in Classical Planning
TLDR
This work introduces a partial grounding approach that grounds only a projection of the task, when complete grounding is not feasible, and proposes a guiding mechanism that, for a given domain, identifies the parts of a task that are relevant to find a plan by using off-the-shelf machine learning methods.
Learning Generalized Reactive Policies using Deep Neural Networks
TLDR
This work shows that a deep neural network can be used to learn and represent a generalized reactive policy (GRP) that maps a problem instance and a state to an action, and that the learned GRPs efficiently solve large classes of challenging problem instances.
Learning Domain-Independent Planning Heuristics with Hypergraph Networks
TLDR
This work presents the first approach capable of learning domain-independent planning heuristics entirely from scratch, and shows that the heuristically learned are able to generalise across different problems and domains, including to domains that were not seen during training.
Width and Serialization of Classical Planning Problems
TLDR
A width parameter is introduced that bounds the complexity of classical planning problems and domains, along with a simple but effective blind-search procedure that runs in time that is exponential in the problem width, resulting in a 'blind' planner that competes well with a best-first search planner guided by state-of-the-art heuristics.
Learning Control Knowledge for Forward Search Planning
TLDR
This work introduces a novel feature space for representing control knowledge in terms of information computed via relaxed plan extraction, which has been a major source of success for non-learning planners and gives a new way of leveraging relaxed planning techniques in the context of learning.
Relational envelope-based planning
TLDR
This work shows how structured representations of the environment's dynamics can constrain and speed up the planning process and uses the envelope MDP framework to create a Markov decision process out of a subset of the possible state space.
Learning Classical Planning Strategies with Policy Gradient
TLDR
This paper introduces a novel search framework capable of alternating between several forward search approaches while solving a particular planning problem, and shows that the learner is able to discover domain-specific search strategies.
The Fast Downward Planning System
  • M. Helmert
  • Computer Science
    J. Artif. Intell. Res.
  • 2006
TLDR
A full account of Fast Downward's approach to solving multivalued planning tasks is given and a new non-heuristic search algorithm called focused iterative-broadening search, which utilizes the information encoded in causal graphs in a novel way is presented.
Abstraction and Approximate Decision-Theoretic Planning
...
1
2
3
4
5
...