Cooperative, Dynamics-based, and Abstraction-Guided Multi-robot Motion Planning

Authors: Duong Le, Erion Plaku

JAIR 2018 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments using vehicle models with nonlinear dynamics operating in complex environments, where cooperation among robots is required, show significant speedups over related work. Experiments are conducted using 6 scenes, as shown in Figure 1 and 4. Figure 5 shows the results of the comparisons.
Researcher Affiliation Academia Duong Le EMAIL Erion Plaku EMAIL Catholic University of America Department of Electrical Engineering and Computer Science Washington, DC 20064 USA
Pseudocode Yes Algorithm 1 Pseudocode for the construction of the roadmap Mi = (VMi, EMi, cost Mi) Algorithm 2 Pseudocode for the proposed approach
Open Source Code No The paper mentions 'Videos of solutions obtained by our approach on this and other scenes can be found at http://goo.gl/8muxw C. Figure best viewed in color and on screen.' This is a link to videos, not source code. There is no explicit statement about making the source code available.
Open Datasets No For each scene and number of robots n a set of 60 problem instances is generated, denoted by I scene,n , by randomly placing the robots and the goals in the environment. To make the test cases more challenging, rather than randomly sampling from the entire W, the robots and the goals are placed at random locations inside certain manually-selected areas. The paper describes creating custom problem instances but does not provide concrete access information (link, DOI, repository) for them to be publicly available.
Dataset Splits No For a scene and number of robots n, results report the mean runtime obtained over the 60 problem instances in I scene,n , after dropping the best and worst five runs to avoid the influence of outliers. This describes a method for selecting runs for evaluation from the generated instances, but it does not specify explicit training, validation, or test dataset splits in the typical sense for model training and evaluation.
Hardware Specification Yes Experiments were run on an Intel Core i7 (1.90GHz).
Software Dependencies No The paper mentions using Bullet (Coumans, 2012) and ODE (Smith, 2006) as physics game engines, PQP (Larsen et al., 1999), FCL (Pan et al., 2012) for collision checking, and algorithms like WHCA*(Silver, 2005), SIPP (Narayanan et al., 2012), and Push-and-Swap (Luna & Bekris, 2011). However, it does not provide specific version numbers for these software components as used in the authors' implementation, nor for any other core libraries or programming languages.
Experiment Setup Yes The motion equations fi are often expressed as a set of differential equations of the form s = fi(s, a), (1), where s Si, a Ai, and s is the derivative of s. simulate(s, a, fi, dt), which computes the new robot state snew Si, obtained by starting at the state s Si and applying the control action a Ai for a time step dt R 0. A proportional-integrative-derivative (PID) controller (Spong, Hutchinson, & Vidyasagar, 2005) is used to select the control actions that steer robot Ri toward ctarget i. The weight for Γ c1,...,cn is defined as w(Γ c1,...,cn ) = αNr Sel(Γ c1,...,cn ) / (Σ Pn i=1(cost(σi))2), where 0 < α < 1. Experiments were run using WHCA* with three different window sizes (2, 5, and 10) (Silver, 2005), SIPP (Narayanan et al., 2012), and Push-and-Swap (Luna & Bekris, 2011).