Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
KABouM: Knowledge-Level Action and Bounding Geometry Motion Planner
Authors: Andre Gaschler, Ronald P. A. Petrick, Oussama Khatib, Alois Knoll
JAIR 2018 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our approach on a wide set of problems using real robots, including tasks with multiple manipulators, sensing and branched plans, and mobile manipulation. |
| Researcher Affiliation | Academia | Andre Gaschler EMAIL fortiss An-Institut Technische Universit at M unchen Munich, Germany Ronald P. A. Petrick EMAIL Department of Computer Science, Heriot-Watt University Edinburgh, United Kingdom Oussama Khatib EMAIL Artificial Intelligence Laboratory, Stanford University Stanford, USA Alois Knoll EMAIL Institut f ur Informatik, Technische Universit at M unchen Garching b. M unchen, Germany |
| Pseudocode | Yes | Algorithm 1 Bounded edge contraction point optimisation Input: Neighbouring planes P of an edge Output: Edge contraction point v and its cost... Algorithm 2 Bounding mesh generation with iterative edge contraction Input: Mesh m, tolerance ε Output: Bounding mesh m m |
| Open Source Code | No | The domain definitions of the evaluated scenarios will be made available on the first author’s website http://www.andre-gaschler.de/ at the time of publication. |
| Open Datasets | No | In the Remove n Objects scenario,... problem instances were generated systematically: random locations were evenly distributed over a rectangular table area until a non-colliding set was found, locations in a line were generated with a defined starting position and increment. |
| Dataset Splits | No | problem instances were generated systematically: random locations were evenly distributed over a rectangular table area until a non-colliding set was found, locations in a line were generated with a defined starting position and increment. |
| Hardware Specification | Yes | All plans were generated on desktop computers with a dual-core 2.8 GHz processor and 8 GB of memory. |
| Software Dependencies | No | High-level planning capabilities in KABou M are provided by the Planning with Knowledge and Sensing system (PKS, Petrick & Bacchus, 2002, 2004), an existing general-purpose planner |
| Experiment Setup | Yes | In the Bartender scenario, the robot CAD models (86,746 vertices) are first simplified to bounding meshes of 9,030 vertices, and then decomposed into only 22 convex bodies with a total of 963 vertices, all at a bounded geometric error ε < 0.03 m. |