Arena-Bench: A Benchmarking Suite for Obstacle Avoidance Approaches in Highly Dynamic Environments

Abstract

The ability to autonomously navigate safely, especially within dynamic environments, is paramount for mobile robotics. In recent years, DRL approaches have shown superior performance in dynamic obstacle avoidance. However, these learning-based approaches are often developed in specially designed simulation environments and are hard to test against conventional planning approaches. Furthermore, the integration and deployment of these approaches into real robotic platforms are not yet completely solved. In this paper, we present Arenabench, a benchmark suite to train, test, and evaluate navigation planners on different robotic platforms within 3D environments. It provides tools to design and generate highly dynamic evaluation worlds, scenarios, and tasks for autonomous navigation and is fully integrated into the robot operating system. To demonstrate the functionalities of our suite, we trained a DRL agent on our platform and compared it against a variety of existing different model-based and learning-based navigation approaches on a variety of relevant metrics. Finally, we deployed the approaches towards real robots and demonstrated the reproducibility of the results. The code is publicly available at github.com/ignc-research/arena-bench.

Arena-Rosnav 2.0: A Development and Benchmarking Platform for Robot Navigation in Highly Dynamic Environments

“…This work extends our previous works Arena-Bench [1] and Arena-Rosnav [2], where a platform for developing and training DRL agents within a resource-efficient 2D environment was proposed and tools were provided to benchmark the trained agent against several model-based and learning based approaches on different highly dynamic scenarios respectively. DRL based approaches have shown remarkable results for navigation in highly dynamic environments and a variety of research works incorporated DRL into their systems [4], [5], [6] [7], [8], [9].…”

“…In recent years, Deep Reinforcement Learning (DRL) for navigation in dynamic environments has achieved remarkable results and was applied by a variety of researchers. However, several works have pointed out downsides and challenges when working and developing DRL approaches [3] [1]. In particular, development is still a high hurdle to overcome, the training process is oftentimes difficult and tedious, and comparability with other approaches is not trivial.…”

“…Furthermore, most benchmarks aimed to compare approaches of the same paradigm domain and lack the possibility to compare across the different paradigms. In our previous works, we proposed Arena-Bench [1] to address these issues and provide an unified benchmark, which is able to compare both paradigms side-by-side. Despite a variety of functions and planners, the system had a number of limitations such as the limitation to 2D training and simulation environments, a tedious installation process, and Fig.…”

Deep-Reinforcement-Learning-based Path Planning for Industrial Robots using Distance Sensors as Observation

“…DRL has emerged as an end-to-end approach with the potential to learn complex behavior in unknown environments. In the field of robot navigation, DRL-based approaches have shown promising results [11], [12], [1], [13]. Wen et al propose the use of DDPG to plan the trajectory of a robot arm to realize obstacle avoidance [14] based purely on DRL.…”

“…Similar works by Faust et al [16] and Chiang et al [2] combine DRL-based motion planning with classic approaches such as RRT and PRM for the motion planning of ground robots over long distances. More recently, works by Kästner et al [11], Dugas et al [17], and Guldenring et al [18] showed the superiority of DRL approaches for fast obstacle avoidance in unknown and dynamic environments. DRL-based approaches have also been utilized in a number of research works for motion planning and collision avoidance for stationary robots.…”