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; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
REBA: A Refinement-Based Architecture for Knowledge Representation and Reasoning in Robotics
Authors: Mohan Sridharan, Michael Gelfond, Shiqi Zhang, Jeremy Wyatt
JAIR 2019 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our architecture is evaluated in simulation and on a mobile robot finding and moving target objects to desired locations in indoor domains, to show that the architecture supports reliable and efficient reasoning with violation of defaults, noisy observations and unreliable actions, in complex domains. We evaluate the effect of our representation and use of default knowledge on reliability and computational efficiency of decision making, we first conducted trials in which REBA was compared with REBA , a version that does not include any default knowledge... Figures 8-9 summarize the corresponding results... In addition to the trials in simulated domains, we implemented and evaluated REBA with POMDP-1 on physical robots using the Robot Operating System (ROS). |
| Researcher Affiliation | Academia | Mohan Sridharan EMAIL School of Computer Science University of Birmingham, UK Michael Gelfond EMAIL Department of Computer Science Texas Tech University, USA Shiqi Zhang EMAIL Department of Computer Science SUNY Binghamton, USA Jeremy Wyatt EMAIL School of Computer Science University of Birmingham, UK |
| Pseudocode | Yes | Algorithm 1: Constructing POMDP transition function T P and observation function OP... Algorithm 2: Construction of POMDP reward function RP... Algorithm 3: Control loop of REBA |
| Open Source Code | Yes | Please see example2.sp at https://github.com/mhnsrdhrn/refine-arch for the complete program formalizing this reasoning in SPARC. ... Please see refined.sp at https://github.com/mhnsrdhrn/refine-arch for the ASP program (in SPARC format) describing the refined signature and refined axioms for our illustrative example, along with additional axioms that support planning to achieve particular goals. |
| Open Datasets | No | The paper describes its own custom |
| Dataset Splits | No | The paper describes experimental trials and scenarios (e.g., random object placement, 1000 trials, 10000 trials) but does not provide specific training/validation/test splits for any dataset. |
| Hardware Specification | Yes | Figure 1(b) shows the Peoplebot wheeled robot platform used in these trials. The robot is equipped with a stereo camera, laser range finder, microphone, speaker, and a laptop running Ubuntu Linux that performs all the processing. ... we also conducted 40 experimental trials using the Turtlebot wheeled robot platform in Figure 1(c) in a variant of the illustrative domain in Example 1. This robot was equipped with a Kinect (RGB-D) sensor, a laser range finder, and a laptop running Ubuntu Linux that performs all the processing. |
| Software Dependencies | No | The domain representation for logical reasoning is translated into a program in SPARC ... and the representation for probabilistic reasoning is translated into a partially observable Markov decision process (POMDP) ... on physical robots using the Robot Operating System (ROS). ... existing ROS implementation of a SLAM algorithm (Dissanayake, Newman, & Clark, 2001). The paper mentions software components like SPARC, POMDP solvers, and ROS, but does not provide specific version numbers for any of them. |
| Experiment Setup | Yes | In each experimental trial, the robot s goal was to find and move specific objects to specific places the robot s location, the target object, and locations of domain objects were chosen randomly. ... For the plots corresponding to POMDP-1, the number of actions the robot is allowed to execute before it has to terminate is set to 50. ... discount: 0.99 values: reward ... All non-terminal actions have unit cost. A correct answer receives a large positive reward (100), whereas an incorrect answer receives a large negative reward ( 100). ... Any move from a cell to a neighboring cell succeeds with probability 0.85. ... The grasp action succeeds with probability 0.95; otherwise it fails. ... If the thing being searched for in a cell exists in the cell, 0.95 is the probability of successfully finding it. |