Meta tags:
description=Spatially-grounded parameterized motion primitives enables better geometric reasoning in manipulation.;
keywords=Spatially-grounded motion primitives;

Headings (most frequently used words):

hacman, manipulation, spatially, grounded, motion, primitives, for, rss, 2024, select, object, nbsp, shows, long, horizon, contact, reasoning, generalizing, to, unseen, objects, and, solving, diverse, tasks, abstract, method, overview, robot, experiments, sim, results, with, interactive, visualizations, bibtex,

Text of the page (most frequently used words):
the (28), and (19), primitive (17), for (11), our (8), point (8), #grounded (7), #primitives (7), manipulation (7), hacman (6), motion (6), object (6), with (6), parameters (6), each (6), robot (5), method (5), spatially (4), pose (4), visualize (3), select (3), set (3), parameterized (3), location (3), where (3), will (3), three (3), components (3), action (3), policy (3), learning (3), long (3), horizon (3), grounding (3), generalization (3), tasks (3), from (2), jiang (2), bowen (2), yilin (2), zhou (2), wenxuan (2), paxton (2), chris (2), held (2), david (2), 2024 (2), cup (2), mug (2), interactive (2), sim (2), cloud (2), estimate (2), per (2), a_i (2), x_i (2), critic (2), one (2), value (2), during (2), consisting (2), type (2), these (2), form (2), space (2), that (2), reinforcement (2), task (2), shown (2), bin (2), spatial (2), learned (2), end (2), variations (2), such (2), grasp (2), push (2), contact (2), world (2), diverse (2), reasoning (2), real (2), unseen (2), objects (2), website, template, borrowed, nerfies, inproceedings, jiang2024hacmanpp, title, author, journal, robotics, science, systems, year, bibtex, bottle, plant, container, drag, slider, different, timesteps, click, legends, show, hide, elements, plot, visuliazation, visualizations, results, rubik, cube, bowl, tennis, ball, box, car, nbsp, experiments, processes, then, compute, maps, which, using, estimated, either, sample, map, training, choose, highest, score, evaluation, execution, consists, library, emph, left, train, learns, sequence, their, corresponding, locations, perform, here, placed, initial, must, moved, into, second, target, top, selected, selecting, heatmaps, overview, although, has, some, success, policies, are, often, not, sufficiently, robust, geometry, improve, introduce, specifically, propose, representation, what, execute, gripper, make, how, executed, specifying, direction, orientation, define, novel, discrete, continuous, framework, enables, agents, learn, chain, together, appropriate, complete, environment, able, effectively, generalize, across, shape, approach, significantly, outperforms, existing, methods, particularly, complex, scenarios, demanding, both, high, level, sequential, zero, shot, transfer, succeeds, challenging, abstract, shows, generalizing, solving, code, arxiv, paper, equal, contribution, meta, carnegie, mellon, university, rss, more, research,

Text of the page (random words):
hacman more research hacman hacman spatially grounded motion primitives for manipulation rss 2024 bowen jiang 1 yilin wu 1 wenxuan zhou 1 chris paxton 2 david held 1 1 carnegie mellon university 2 meta ai equal contribution paper arxiv code hacman shows long horizon contact reasoning generalizing to unseen objects and solving diverse manipulation tasks abstract although end to end robot learning has shown some success for robot manipulation the learned policies are often not sufficiently robust to variations in object pose or geometry to improve the policy generalization we introduce spatially grounded parameterized motion primitives specifically we propose an action representation consisting of three components what primitive type such as grasp or push to execute where the primitive will be grounded e g where the gripper will make contact with the world and how the primitive motion is executed such as parameters specifying the push direction or grasp orientation these three components define a novel discrete continuous action space for reinforcement learning our framework enables robot agents to learn to chain diverse motion primitives together and select appropriate primitive parameters to complete long horizon manipulation tasks by grounding the primitives on a spatial location in the environment our method is able to effectively generalize across object shape and pose variations our approach significantly outperforms existing methods particularly in complex scenarios demanding both high level sequential reasoning and object generalization with zero shot sim to real transfer our policy succeeds in challenging real world manipulation tasks with generalization to unseen objects method overview our method consists of a library of parameterized emph spatially grounded motion primitives left consisting of a primitive type primitive location where the primitive will be grounded and primitive parameters these three components form the action space for a policy that we train with reinforcement learning our method learns to select a sequence of primitives and their corresponding locations and parameters to perform a long horizon manipulation task in the task shown here the object is placed in one bin in an initial pose and it must be moved into a second bin in a target pose at the top we visualize the spatial grounding for the selected primitive for each point we visualize the learned q value of selecting that point in the form of heatmaps as the grounding location for each primitive our method processes a point cloud to estimate a set of per point primitive parameters a_i m for each point x_i in the point cloud and for each primitive in our primitive set we then compute a set of critic maps one per primitive which estimate the q value q_ i k of using each primitive k grounded at each point x_i and parameterized by the estimated primitive parameters a_i m we either sample from the critic map during training or choose the point and primitive with the highest score during evaluation for robot execution robot experiments select object nbsp car box cup tennis ball bowl rubik s cube sim results with interactive visualizations interactive visuliazation drag the slider to visualize different timesteps click on the legends on the plot to show hide elements object mug mug 2 cup plant container bottle bibtex inproceedings jiang2024hacmanpp title hacman spatially grounded motion primitives for manipulation author jiang bowen and wu yilin and zhou wenxuan and and paxton chris and held david journal robotics science and systems year 2024 website template borrowed from nerfies and hacman