Towards Stable Symbol Grounding with Zero-Suppressed State AutoEncoder

Asai, Masataro; Kajino, Hiroshi

doi:10.1609/icaps.v29i1.3526

Cited by 3 publications

(1 citation statement)

References 17 publications

(31 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Another approach to reducing reliance on state labelling is to provide alternative information that is easier to access, such as state images. A significant contribution in this area is Latplan (Asai and Fukunaga 2018;Asai and Kajino 2019;Asai and Muise 2020;Asai et al 2022), an unsupervised neuro-symbolic model, based on an auto-encoder framework that exclusively utilizes state images to recover action models. While both Latplan and our approach apply neuro-symbolic methods to state images, there are significant distinctions between the two works, leading to complementary strengths and weaknesses.…”

Section: Related Workmentioning

confidence: 99%

Neuro-Symbolic Learning of Lifted Action Models from Visual Traces

Xi,

Gould,

Thiébaux

2024

ICAPS

View full text Add to dashboard Cite

Model-based planners rely on action models to describe available actions in terms of their preconditions and effects. Nonetheless, manually encoding such models is challenging, especially in complex domains. Numerous methods have been proposed to learn action models from examples of plan execution traces. However, high-level information, such as state labels within traces, is often unavailable and needs to be inferred indirectly from raw observations. In this paper, we aim to learn lifted action models from visual traces --- sequences of image-action pairs depicting discrete successive trace steps. We present ROSAME, a differentiable neuRO-Symbolic Action Model lEarner that infers action models from traces consisting of probabilistic state predictions and actions. By combining ROSAME with a deep learning computer vision model, we create an end-to-end framework that jointly learns state predictions from images and infers symbolic action models. Experimental results demonstrate that our method succeeds in both tasks, using different visual state representations, with the learned action models often matching or even surpassing those created by humans.

show abstract

Section: Related Workmentioning

confidence: 99%

Neuro-Symbolic Learning of Lifted Action Models from Visual Traces

Xi,

Gould,

Thiébaux

2024

ICAPS

View full text Add to dashboard Cite

show abstract

Planning from Pixels in Atari with Learned Symbolic Representations

Dittadi

Drachmann

Bolander

2021

AAAI

View full text Add to dashboard Cite

Width-based planning methods have been shown to yield state-of-the-art performance in the Atari 2600 domain using pixel input. One successful approach, RolloutIW, represents states with the B-PROST boolean feature set. An augmented version of RolloutIW, pi-IW, shows that learned features can be competitive with handcrafted ones for width-based search. In this paper, we leverage variational autoencoders (VAEs) to learn features directly from pixels in a principled manner, and without supervision. The inference model of the trained VAEs extracts boolean features from pixels, and RolloutIW plans with these features. The resulting combination outperforms the original RolloutIW and human professional play on Atari 2600 and drastically reduces the size of the feature set.

show abstract

Classical Planning in Deep Latent Space

Asai¹,

Kajino

Fukunaga

et al. 2022

jair

View full text Add to dashboard Cite

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), Latplan learns a complete propositional PDDL action model of the environment. Later, when a pair of images representing the initial and the goal states (planning inputs) is given, Latplan finds a plan to the goal state in a symbolic latent space and returns a visualized plan execution. We evaluate Latplan using image-based versions of 6 planning domains: 8-puzzle, 15-Puzzle, Blocksworld, Sokoban and Two variations of LightsOut.

show abstract

Towards Stable Symbol Grounding with Zero-Suppressed State AutoEncoder

Cited by 3 publications

References 17 publications

Neuro-Symbolic Learning of Lifted Action Models from Visual Traces

Neuro-Symbolic Learning of Lifted Action Models from Visual Traces

Planning from Pixels in Atari with Learned Symbolic Representations

Classical Planning in Deep Latent Space

Contact Info

Product

Resources

About