Learning Symbolic Operators for Task and Motion Planning

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

doi:10.1109/iros51168.2021.9635941

Cited by 35 publications

(40 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Comparing Ours to Down Eval shows that assuming downward refinability at evaluation time works for PickPlace1D, Blocks, and Painting, but not for Tools. We were surprised by this result because the manually designed abstractions for PickPlace1D and Painting are not downward refinable [34]. Upon inspection, we find that Ours learns abstractions that are downward refinable.…”

Section: Resultsmentioning

confidence: 91%

“…For PickPlace1D, the learned "target clear" predicate leads to downward refinability, as discussed with P4 in the previous paragraph. For Painting, a learned "box lid open" predicate resolves the downward refinability issue discussed in prior work [34], where the position of the box lid (open or closed) was not modeled in the manual abstraction. In contrast, the abstractions learned by Ours for the Tools environment are not downward refinable; for example, it is not possible to determine whether a screwdriver's shape is compatible with that of a screw, at the abstract level.…”

Section: Resultsmentioning

confidence: 99%

“…We now briefly describe the environments, with further details (including state and action spaces) in Appendix A.2. The first three environments were established in prior work by Silver et al [34], but in that work, all state abstractions were manually defined (we use the same state abstractions for our Manual baseline below).…”

Section: Methodsmentioning

confidence: 99%

“…In the course of our research, we considered many options inspired by prior work, including per-operator prediction error [12,34], bisimulation [17,29], and inverse planning-based objectives [35,36], but found them all to be divergent from J real , leading to poor performance (see the baselines in Section 6). Our main insight was based on the following observation: although sampler learning and REFINE are slow, operator learning and abstract search (on the training tasks) are both fast -operator learning takes time linear in the size of the dataset, and abstract search is guided by powerful AI planning heuristics.…”

Section: Scoring a Candidate Predicate Setmentioning

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: Resultsmentioning

confidence: 91%

Section: Resultsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Scoring a Candidate Predicate Setmentioning

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

“…To speed up the planning process in TAMP and increase its probability of finding an overall feasible solution in a given time-budget, several heuristics have been introduced to reduce the search space. In particular, in the era of robot learning, different learned heuristics have been integrated in TAMP to accelerate its decision-making process, and support its probabilistic completeness to find a feasible plan in a specific planning horizon [6], [7], [8]. [7] introduces a more general visualbased neural network to do the classification of tabletop manipulation.…”

Section: Introductionmentioning

confidence: 99%

Accelerating Integrated Task and Motion Planning with Neural Feasibility Checking

Xu¹,

Ren²,

Chalvatzaki³

et al. 2022

Preprint

View full text Add to dashboard Cite

As robots play an increasingly important role in the industrial, the expectations about their applications for everyday living tasks are getting higher. Robots need to perform long-horizon tasks that consist of several sub-tasks that need to be accomplished. Task and Motion Planning (TAMP) provides a hierarchical framework to handle the sequential nature of manipulation tasks by interleaving a symbolic task planner that generates a possible action sequence, with a motion planner that checks the kinematic feasibility in the geometric world, generating robot trajectories if several constraints are satisfied, e.g., a collision-free trajectory from one state to another. Hence, the reasoning about the task plan's geometric grounding is taken over by the motion planner. However, motion planning is computationally intense and is usability as feasibility checker casts TAMP methods inapplicable to real-world scenarios. In this paper, we introduce neural feasibility classifier (NFC), a simple yet effective visual heuristic for classifying the feasibility of proposed actions in TAMP. Namely, NFC will identify infeasible actions of the task planner without the need for costly motion planning, hence reducing planning time in multistep manipulation tasks. NFC encodes the image of the robot's workspace into a feature map thanks to convolutional neural network (CNN). We train NFC using simulated data from TAMP problems and label the instances based on IK feasibility checking. Our empirical results in different simulated manipulation tasks show that our NFC generalizes to the entire robot workspace and has high prediction accuracy even in scenes with multiple obstructions. When combined with state-of-theart integrated TAMP, our NFC enhances its performance while reducing its planning time.

show abstract