Learning Neuro-Symbolic Relational Transition Models for Bilevel Planning

Chitnis, Rohan; Silver, Tom; Tenenbaum, Joshua B.; Lozano-Pérez, Tomás; Kaelbling, Leslie Pack

doi:10.48550/arxiv.2105.14074

Cited by 4 publications

(9 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…• Dense subgoal sequence. The expected abstract state sequence serves as a dense subgoal sequence for the REFINE procedure, in the sense that if there is ever a deviation from it, REFINE is able to resample and/or backtrack immediately (rather than, say, needing to roll out trajectories up to the end of the time horizon before testing the goal [20]).…”

Section: Discussion: the Virtues Of Abstractions In Bilevel Planningmentioning

confidence: 99%

“…Our operator learning method is largely based on prior work [20,27,28]. We describe it here briefly and refer the reader to Appendix A.1 for an extended example.…”

Section: Learning Operatorsmentioning

confidence: 99%

“…We use these substitutions for the third and final step of operator learning: precondition learning. For this, we perform an intersection over all abstract states in each equivalence class [17,20,29]:…”

Section: Learning Operatorsmentioning

confidence: 99%

“…The agent knows the transition model of the environment, but it is highly intractable to plan directly using this model in the continuous spaces, motivating the need to learn abstractions for guidance. To obtain data-efficient generalization over object identities, we learn relational, neuro-symbolic abstractions, where the symbolic components are predicates and operators, like those used in AI planning [11], and the neural components are samplers that refine the abstractions into actions that can be executed in the actual environment [18,19,20].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Inventing Relational State and Action Abstractions for Effective and Efficient Bilevel Planning

Silver¹,

Chitnis²,

Kumar³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

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 perform bilevel planning with abstractions, where a high-level search for abstract plans is used to guide planning in the original transition space. In this paper, we develop a novel framework for learning state and action abstractions that are explicitly optimized for both effective (successful) and efficient (fast) bilevel planning. Given demonstrations of tasks in an environment, our data-efficient approach learns relational, neuro-symbolic abstractions that generalize over object identities and numbers. The symbolic components resemble the STRIPS predicates and operators found in AI planning, and the neural components refine the abstractions into actions that can be executed in the environment. Experimentally, we show across four robotic planning environments that our learned abstractions are able to quickly solve held-out tasks of longer horizons than were seen in the demonstrations, and can even outperform the efficiency of abstractions that we manually specified. We also find that as the planner configuration varies, the learned abstractions adapt accordingly, indicating that our abstraction learning method is both "task-aware" and "planner-aware." Code: https://tinyurl.com/predicators-release

show abstract

Section: Discussion: the Virtues Of Abstractions In Bilevel Planningmentioning

confidence: 99%

“…Our operator learning method is largely based on prior work [20,27,28]. We describe it here briefly and refer the reader to Appendix A.1 for an extended example.…”

Section: Learning Operatorsmentioning

confidence: 99%

Section: Learning Operatorsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Inventing Relational State and Action Abstractions for Effective and Efficient Bilevel Planning

Silver¹,

Chitnis²,

Kumar³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Our approach relates to different uses of neuro-symbolic architectures in other fields such as analysis of images [33][34][35] and videos [36,37]. Recent work by Chitnis et al [38] uses symbolic planning to guide the continuous planning process. These architectures combine neural components with symbolic representations such as programs or domain specific languages.…”

Section: Neuro-symbolic Network and Architecturesmentioning

confidence: 99%

Trajectory Prediction with Linguistic Representations

Kuo¹,

Huang²,

Barbu³

et al. 2021

Preprint

View full text Add to dashboard Cite

Language allows humans to build mental models that interpret what is happening around them resulting in more accurate long-term predictions. We present a novel trajectory prediction model that uses linguistic intermediate representations to forecast trajectories, and is trained using trajectory samples with partially-annotated captions. The model learns the meaning of each of the words without direct per-word supervision. At inference time, it generates a linguistic description of trajectories which captures maneuvers and interactions over an extended time interval. This generated description is used to refine predictions of the trajectories of multiple agents. We train and validate our model on the Argoverse dataset, and demonstrate improved accuracy results in trajectory prediction. In addition, our model is more interpretable: it presents part of its reasoning in plain language as captions, which can aid model development and can aid in building confidence in the model before deploying it.

show abstract

Generalizable Task Planning through Representation Pretraining

Wang¹,

Xu²,

Li³

2022

Preprint

View full text Add to dashboard Cite

The ability to plan for multi-step manipulation tasks in unseen situations is crucial for future home robots. But collecting sufficient experience data for end-to-end learning is often infeasible in the real world, as deploying robots in many environments can be prohibitively expensive. On the other hand, large-scale scene understanding datasets contain diverse and rich semantic and geometric information. But how to leverage such information for manipulation remains an open problem. In this paper, we propose a learning-to-plan method that can generalize to new object instances by leveraging object-level representations extracted from a synthetic scene understanding dataset. We evaluate our method with a suite of challenging multi-step manipulation tasks inspired by household activities [28] and show that our model achieves measurably better success rate than state-of-the-art end-to-end approaches. Additional information can be found at: https: //sites.google.com/view/gentp

show abstract

Learning Neuro-Symbolic Relational Transition Models for Bilevel Planning

Cited by 4 publications

References 32 publications

Inventing Relational State and Action Abstractions for Effective and Efficient Bilevel Planning

Inventing Relational State and Action Abstractions for Effective and Efficient Bilevel Planning

Trajectory Prediction with Linguistic Representations

Generalizable Task Planning through Representation Pretraining

Contact Info

Product

Resources

About