Hierarchical task and motion planning in the now

By Leslie Pack Kaelbling, Tomas Lozano-Perez. 2011

680 citations recorded by the source corpus.

In this paper we outline an approach to the integration of task planning and motion planning that has the following key properties: It is aggressively hierarchical; it makes choices and commits to them in a top-down fashion in an attempt to limit the length of plans that need to be constructed, and thereby exponentially decrease the amount of search required. It operates on detailed, continuous geometric representations and does not require a-priori discretization of the state or action spaces.

Subfields: Aerospace Engineering, 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 Aerospace Engineering

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Reinforcement Learning

Citations in the corpus

ALFWorld: Aligning Text and Embodied Environments for Interactive Learning (2020, background): “Using symbolic representations for state and action is also an inherent characteristic of works in task-and-motionplanning (Kaelbling and Lozano-Pérez, 2011; Konidaris et al., 2018) and symbolic planning (Asai and Fukunaga, 2017).”
Value Iteration Networks (2016, methodology): “In future work we intend to learn different planning computations, based on simulation [10], or optimal linear control [30], and combine them with reactive policies, to potentially develop RL solutions for task and motion planning [14].”
Overcoming Exploration in Reinforcement Learning with Demonstrations (2017, background): “More recent work on task and motion planning considers both logical and physical aspects of the task [27], [28], [29], but requires domainspecific engineering.”
IQA: Visual Question Answering in Interactive Environments (2017, background): “A large body of research on planning algorithms [12, 14, 26, 62, 63] use high-level formal languages.”
Skill Induction and Planning with Latent Language (2021, background): “…specifications, limiting their 045 applicability in complex and hard-to-formalize en- 046 vironments, or are restricted to extremely simple 047 high-level actions, limiting their expressive power 048 (Bacon et al., 2017; Sutton et al., 1999; Dietterich, 049 1999; Kaelbling and Lozano-Pérez, 2011).”
Visual Semantic Planning Using Deep Successor Representations (2017, background): “Task-level planning [10, 13, 20, 47, 48] addresses the problem of finding a high-level plan for performing a task.”
Differentiable Physics and Stable Modes for Tool-Use and Manipulation Planning (2018, methodology): “the solver could be improved by combining LGP with more efficient AI planners [11, 9].”
Integrated Task and Motion Planning (2020, neutral): “Interleaved Dornhege∗ (62, 63, 91) Gaschler∗ (92, 93, 94) Colledanchise∗ (95) Gravot∗ (96, 97) Stilman† (23, 98, 99) Plaku† (100) Kaelbling∗ (101, 102) Barry† (103, 30, 104) Garrett∗ (70, 71) Thomason∗ (105) Kim∗ (106, 107) Kingston† (108) Fernandez-Gonzalez∗ (109)”
Dynamics Learning with Cascaded Variational Inference for Multi-Step Manipulation (2019, methodology): “Conventional methods have formulated this as the task and motion planning problem [3, 4].”
Regression Planning Networks (2019, methodology): “Following [12], we define goal g ∈ G as the conjunction of a set of logical atoms, each consists of a predicate and a list of object arguments, e.”
Text2Motion: from natural language instructions to feasible plans (2023, background): “For instance, Language-conditioned policies (LCPs) are not only used to learn short-horizon skills [9–14], but also long-horizon policies [15–17].”
Neural Task Graphs: Generalizing to Unseen Tasks From a Single Video Demonstration (2018, background): “Other representations include: integrating task and motion planning [21] and behavior-based systems [30].”
Model-based hierarchical reinforcement learning and human action control (2014, background): “If I decide I want to go to a particular cash-only restaurant for dinner but realize that my wallet is empty, knowledge of outcomes allows me to see immediately that a trip to the ATM should be added to my plan (see [24]).”
Deliberation for autonomous robots: A survey (2017, methodology): “Asymov [36] CPMP [44] I-TMP* [97] CTAMP [22] SamplSGD [238] Hierarchical AHP [231] Kaelbling [113] Table 5: Integrated motion and task planning, (*) with online planning and acting Motion and manipulation planning are key capabilities for a robot, requiring speciﬁc representations for geometry,…”
Physically Embedded Planning Problems: New Challenges for Reinforcement Learning (2020, methodology): “, 2018), and progress in this area has largely been driven by the design and engineering of domain-specific hand-crafted and structured methods (Dantam et al., 2016; Kaelbling and Lozano-Pérez, 2011; Plaku and Hager, 2010; Siméon et al., 2004; Srivastava et al., 2014; Toussaint, 2015), which typical”
Integrated task and motion planning in belief space (2013, background): “…that combines geometric and symbolic representations in an aggressively hierarchical planning architecture, called HPN (for hierarchical planning in the now), which we summarize in Section 1.2 and discuss in detail in Kaelbling and Lozano-Pérez (2011) and Kaelbling and Lozano-Pérez (2012a).”
Modeling Long-horizon Tasks as Sequential Interaction Landscapes (2020, background): “Hierarchical and Symbolic Planning: it has been recognized that abstract symbols can serve as a meaningful representation to structure tasks [7, 49] and to organize their often hierarchical properties [24].”
Learning Interactive Real-World Simulators (2023, neutral): “• Panorama scans.”