Classical Planning in Deep Latent Space: Bridging the Subsymbolic-Symbolic Boundary
By Masataro Asai, Alex S. Fukunaga. 2017
191 citations recorded by the source corpus.
Current domain-independent, classical planners require symbolic models of the problem domain and instance as input, resulting in a knowledge acquisition bottleneck. Meanwhile, although deep learning has achieved significant success in many fields, the knowledge is encoded in a subsymbolic representation which is incompatible with symbolic systems such as planners. We propose LatPlan, an unsupervised architecture combining deep learning and classical planning. Given only an unlabeled set of image pairs showing a subset of transitions allowed in the environment (training inputs), and a pair of images representing the initial and the goal states (planning inputs), LatPlan finds a plan to the goal state in a symbolic latent space and returns a visualized plan execution. The contribution of this paper is twofold: (1) State Autoencoder, which finds a propositional state representation of the environment using a Variational Autoencoder. It generates a discrete latent vector from the images, based on which a PDDL model can be constructed and then solved by an off-the-shelf planner. (2) Action Autoencoder / Discriminator, a neural architecture which jointly finds the action symbols and the implicit action models (preconditions/effects), and provides a successor function for the implicit graph search. We evaluate LatPlan using image-based versions of 3 planning domains: 8-puzzle, Towers of Hanoi and LightsOut.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Reinforcement Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Dieter Fox
- Sergey Levine
- Josh Tenenbaum
- Roozbeh Mottaghi
- Yonatan Bisk
- Ali Farhadi
- Luke S. Zettlemoyer
- Chelsea Finn
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Li Fei-Fei
- Jitendra Malik
- Trevor Darrell
- Kristen Grauman
- Cordelia Schmid
- Dacheng Tao
- Abhinav Kumar Gupta
In Reinforcement Learning
- Dieter Fox
- Josh Tenenbaum
- Sergey Levine
- Roozbeh Mottaghi
- Mohit Shridhar
- Hector Geffner
- Daniel Gordon
- Blai Bonet
Citations in the corpus
- ALFRED: A Benchmark for Interpreting Grounded Instructions for Everyday Tasks (2019, background): “These challenges may be approachable by models that exploit hierarchy [8, 26], modularity [4, 16], and structured reasoning and planning [6].”
- Learning First-Order Symbolic Representations for Planning from the Structure of the State Space (2019, methodology): “An alternative approach produces first-order representations using a class of variational autoencoders that provide a lowdimensional encoding of the images representing the states [4, 3].”
- Unsupervised Grounding of Plannable First-Order Logic Representation from Images (2019, background): “In contrast, Latplan (Asai and Fukunaga 2018) is a framework for domain-independent image-based classical planning.”
- Grounding Predicates through Actions (2021, background): “Some works attempt to learn the symbols themselves, so that perception and planning can work together in an end-to-end fashion [2, 26].”
- Inventing Relational State and Action Abstractions for Effective and Efficient Bilevel Planning (2022, background): “Most relevant are works that learn symbolic state and action abstractions compatible with AI planners [13, 28, 29, 53, 54, 55, 56, 57, 58].”