Right Place, Right Time: Proactive Multi-Robot Task Allocation Under Spatiotemporal Uncertainty
Authors: Charlie Street, Bruno Lacerda, Manuel Mühlig, Nick Hawes
JAIR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we demonstrate the efficacy of our framework across multiple problems in simulation by comparing against a number of baselines. We simulate robot behaviour using the context-aware multi-agent simulator (CAMAS), which captures the task-level behaviour of an MRS acting on a topological map (Street et al., 2022a). CAMAS samples through a multi-robot Markov automaton that uses continuous distributions over the duration of topological edges, and explicitly captures the effects of physical robot interactions such as congestion. All experiments are run on Ubuntu 18.04, with an Intel Core i9-10900K CPU@3.7GHz and 32GB of RAM. All software is written in Python, except for PRISM (Kwiatkowska et al., 2011), which we use for task CTMC analysis, which is written in Java/C++. |
| Researcher Affiliation | Collaboration | Charlie Street EMAIL School of Computer Science, University of Birmingham, UK Bruno Lacerda EMAIL Oxford Robotics Institute, University of Oxford, UK Manuel M uhlig EMAIL Honda Research Institute Europe Gmb H, Offenbach, Germany Nick Hawes EMAIL Oxford Robotics Institute, University of Oxford, UK |
| Pseudocode | No | The paper describes methods and processes in text and flowcharts (Figure 4) but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks with structured code-like formatting. |
| Open Source Code | No | The paper mentions software used, such as CAMAS and PRISM, but does not provide specific access information (e.g., repository links or explicit statements of code release by the authors for their own methodology). |
| Open Datasets | No | The paper mentions using a 'supermarket' and a '10x10 grid world' as experimental environments. It states, 'Duration distributions for the grid world are synthetic, and fitted using moment matching techniques (Marie, 1980). For the supermarket, we use a Gazebo simulation (Koenig & Howard, 2004) of the environment, and use edge partitioning to collect navigation duration data, as presented by Street et al. (2022a).' This describes how data was generated or collected for their simulation but does not provide concrete access information (link, DOI, repository) for a publicly available or open dataset. |
| Dataset Splits | No | The paper describes experimental runs with '40 repeats' but does not specify any training/test/validation dataset splits. The experiments are conducted in simulation environments with randomized parameters, rather than on pre-split datasets. |
| Hardware Specification | Yes | All experiments are run on Ubuntu 18.04, with an Intel Core i9-10900K CPU@3.7GHz and 32GB of RAM. |
| Software Dependencies | Yes | All software is written in Python, except for PRISM (Kwiatkowska et al., 2011), which we use for task CTMC analysis, which is written in Java/C++. |
| Experiment Setup | Yes | In this subsection, we describe our experimental environments, the methods we compare, and the experimental problems for each method. In all problems, a team of robots must service a set of tasks whose announcement is modelled with task CTMCs. The initial robot locations are spread as evenly as possible across the environment, and tasks are serviced instantaneously upon a robot reaching the task location. Robots navigate using a shortest path planner. During bidding, robots use the cheapest insertion heuristic for schedule insertion (Koenig et al., 2006). We consider problems which use the task CTMC structures in Section 4.1, where CTMC transitions are sampled during execution, and all task CTMCs begin in their initial states at the start of execution. For each task CTMC type, we construct problems with 5-20 tasks in increments of 5, where the 10 task problem is the 5 task problem with 5 additional tasks etc. Contiguous and discontiguous sequence CTMCs have three PTLs, as in Fig. 2, and discontiguous star CTMCs have four PTLs, as in Fig. 3. For a given number of tasks, we construct a different problem for each repeat. The expected announcement times, PTLs, and announcement location distributions for each problem are randomised, but kept consistent between methods. PTLs for contiguous sequence CTMCs are constrained to be contiguous across the topological map. |