Decentralized Planning for Car-Like Robotic Swarm in Cluttered Environments Changjia Ma12 Zhichao Han12 Tingrui Zhang12

2025-04-26 0 0 4.55MB 8 页 10玖币

侵权投诉

Decentralized Planning for Car-Like Robotic Swarm in Cluttered

Environments

Changjia Ma1,2, Zhichao Han1,2, Tingrui Zhang1,2,

Jingping Wang1,2, Long Xu1,2, Chengyang Li2, Chao Xu1,2and Fei Gao1,2

Abstract— Robot swarm is a hot spot in robotic research

community. In this paper, we propose a decentralized frame-

work for car-like robotic swarm which is capable of real-time

planning in cluttered environments. In this system, path ﬁnding

is guided by environmental topology information to avoid

frequent topological change, and search-based speed planning is

leveraged to escape from infeasible initial value’s local minima.

Then spatial-temporal optimization is employed to generate

a safe, smooth and dynamically feasible trajectory. During

optimization, the trajectory is discretized by ﬁxed time steps.

Penalty is imposed on the signed distance between agents to

realize collision avoidance, and differential ﬂatness cooperated

with limitation on front steer angle satisﬁes the non-holonomic

constraints. With trajectories broadcast to the wireless network,

agents are able to check and prevent potential collisions. We val-

idate the robustness of our system in simulation and real-world

experiments. Code will be released as open-source packages.

I. INTRODUCTION

Beneﬁting from the ﬂexibility and stability, multi-robot

systems can signiﬁcantly accelerate task completion. Nowa-

days, car-like robotic swarms are widely applied in real life,

such as autonomous trucks in ports, automatic guided vehi-

cles in logistics warehouses, search and rescue in hazardous

scenarios, and exploration in unknown environments. Most

of these systems rely on centralized planning frameworks.

However, with the robot quantity and map complexity in-

creasing, centralized methods suffer from high computation

burden, hence they lack the adaptability to dynamic envi-

ronments, restricting the practical usage of car-like robotic

swarms. For decentralized methods, the computation burden

is separated by each agent. But limited by the performance

of onboard processors, trajectory quality may not be guar-

anteed. To ﬁll this gap, we propose a decentralized planning

framework for car-like robotic swarm which is capable of

real-time planning in obstacle-dense environments.

Different from robots with holonomic kinematics such as

quadrotors, car-like robots with non-holonomic and shape

constraints bring more challenges for online planning in

cluttered environments. During navigation, it is necessary to

conduct frequent replans to avoid collisions with static and

dynamic obstacles. Due to the non-holonomic kinematics,

adjusting to rapid change in trajectories’ topological structure

is more difﬁcult for car-like robots, which results in a higher

rate of colliding with static obstacles, especially when the

agent is close to the obstacle, as shown in Fig.3. To overcome

1State Key Laboratory of Industrial Control Technology, Zhejiang Uni-

versity, Hangzhou 310027, China.(Corresponding author: Fei Gao)

2Huzhou Institute of Zhejiang University, Huzhou 313000, China.

E-mail: {changjiama and fgaoaa}@zju.edu.cn

Fig. 1. Three car-like robots traverse an unexplored environment.

Fig. 2. Car swarm in simulation environment

this difﬁculty, we introduce topology-guided path planning

to guarantee trajectory consistency, which means that the

replanned path remains topology homotopy with the last

planned one.

In the back-end optimization, initial values play an impor-

tant role. Feasible initial values bring beneﬁts such as speed-

ing up optimization convergence and avoiding falling into a

bad solution’s local minima, whereas infeasible initial time

arrangement may lead to convergence to a solution where the

trajectory is in collision with other agents. In order to acquire

reasonable initial time allocation, we introduce search-based

speed planning. Once the spatial path is determined by the

topology-guided path planning, the speed planning generates

a time span that realizes collision avoidance with other agents

as well as dynamic feasibility.

Based on the spatial and temporal initial values, we

employ a spatial-temporal joint optimization to generate

a smooth, safe and feasible trajectory satisfying the non-

holonomic constraints of car-like robots. The proposed sys-

tem is an extension and modiﬁcation of our previous work

[1], which focuses on single-car motion planning. In addition

to the topology-guided path planning and speed planning,

we adopt a ﬁxed time step trajectory discretization strategy

to further improve the computation efﬁciency speciﬁcally

for multi-agent missions. In this system, trajectories are

arXiv:2210.05863v2 [cs.RO] 5 Jul 2023

Fig. 3. The green, blue, yellow and red curves are successively replanned

trajectories. The agent is near the obstacle, but frequent replans lead to the

topological non-homotopy, which makes the red curve intersect with the

obstacle.

broadcast by a communication module, which will be utilized

by other agents to perform real-time planning. The whole

swarm system can traverse unknown environments while

ensuring no collision between agents.

We validate the robustness of our system in a high-

ﬁdelity simulation platform for vehicles [2] and real-world

experiments. Each agent in this swarm independently carries

out online sensing, mapping, planning and control, with

trajectories broadcast to each other. Contributions of this

paper are summarized as below:

1) We introduce topology-guided path planning to im-

prove the motion consistency of the robots, and lever-

age search-based speed planning to provide feasible

initial values for the back-end optimization.

2) We modify the trajectory discretization strategy with

ﬁxed time steps, improving the computation efﬁciency

of the trajectory optimization process.

3) We combine path planning, speed planning with

spatial-temporal optimization, and implement the de-

centralized car-like robotic swarm, realizing real-time

execution in cluttered environments.

4) We will release our code as open-source packages to

serve the research community1.

II. RELATED WORKS

A. Trajectory Optimization for Car-Like Robots

The complexity of kinematic models brings difﬁculties for

planning on car-like robots, and optimization-based methods

have been deeply explored to generate feasible trajectories.

Zhu et al. [3] transform the initial non-convex problem into

a convex optimization with quadratic objective and quadratic

constraints. Zhou et al. [4] use sequential convex optimiza-

tion to relax the curvature constraint. Zhang et al. [5] propose

an optimization-based collision avoidance algorithm. Li et al.

[6] realize collision avoidance by constraining the simpliﬁed

car model in a safety corridor. However, the above meth-

ods still suffer from the complicated mathematical model,

resulting in low computation efﬁciency. Recently, a differ-

ential ﬂatness-based trajectory planning framework [1] for

vehicles is proposed which accomplishes fast computation in

complicated environments. Based on the above baseline [1],

we modiﬁed the trajectory discretization strategy to further

decrease the computation burden in multi-robot scenarios.

1https://github.com/ZJU-FAST-Lab/Car-like-Robotic-swarm

B. Car-like Robotic Swarm in Cluttered Environments

Car-like robotic swarm can be divided as centralized and

decentralized frameworks. Mora et al. [7], [8] realize colli-

sion avoidance based on RVO (Reciprocal Velocity Obsta-

cles) and ORCA (Optimal Reciprocal Collision Avoidance),

and then an optimization process is employed to generate

smooth trajectories. Li et al. [9], [10] solve the time-optimal

MVTP (Multi-Vehicle Trajectory Planning) problems in two

stages and achieve swarm planning in dense environments.

The above frameworks are based on centralized methods,

while centralized methods are stuck by the computation com-

plexity with the increase of the robot number. Delimpaltadaki

et al. [11] propose a decentralized framework to solve the

predecessor-following control problem. However, this work

[11] is limited to platooning scenarios without adaptability to

complicated tasks such as crossovers. In this paper, we adopt

the decentralized planning framework to construct car-like

robotic swarm.

C. Topology-Guided Path Planning

Topological planning has been applied in many scenarios

to choose the best spatial path and avoid local minima. Jaillet

et al. [12] construct visibility deformation roadmaps to deﬁne

the topological homotopy and encode the environmental

topological information. Zhou et al. [13], [14] extend visibil-

ity deformation to propose real-time planning by checking

topological equivalence. Zhou et al. [15] construct distance

ﬁelds in different directions of obstacles and utilize visibility

deformation to ﬁnd distinctive trajectories. In this paper, we

extend visibility deformation in works [12], [13] to a broader

deﬁnition to search for a path homeomorphic with the last

planned path.

D. Speed Planning

Spatial-temporal decomposition is a widely used strategy

in autonomous driving to improve planning efﬁciency, where

speed planning is the temporal part. Works [16], [17] use

search-based speed planning to search for a feasible speed

proﬁle. Works [18], [19] propose optimization-based speed

planning methods to generate a smooth S-T (Space-Time)

curve. Johnson et al. [20], [21] realize dynamic obstacle

avoidance by propagating reachable velocity sets. Other

works [22], [23] conduct speed planning by optimizing

B´

ezier Polynomials. However, most existing methods fo-

cus on getting a smooth and feasible speed proﬁle with-

out spatial-temporal joint optimization. In this paper, we

decouple space-time planning, where initial time span is

provided by the search-based speed planning. Then the initial

path is reﬁned by subsequent spatial-temporal optimization.

Therefore, the initial time span is not necessary to be as

smooth or accurate as existing methods require.

III. PATH AND SPEED PLANNING

In this section, we will introduce topology-guided path

planning and search-based speed planning, which serve

as spatial and temporal initial values for the back-end

optimization.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

DecentralizedPlanningforCar-LikeRoboticSwarminClutteredEnvironmentsChangjiaMa1,2,ZhichaoHan1,2,TingruiZhang1,2,JingpingWang1,2,LongXu1,2,ChengyangLi2,ChaoXu1,2andFeiGao1,2Abstract—Robotswarmisahotspotinroboticresearchcommunity.Inthispaper,weproposeadecentralizedframe-workforcar-likeroboticswarmwhich...

展开>> 收起<<

Decentralized Planning for Car-Like Robotic Swarm in Cluttered Environments Changjia Ma12 Zhichao Han12 Tingrui Zhang12.pdf

共8页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Decentralized Planning for Car-Like Robotic Swarm in Cluttered Environments Changjia Ma12 Zhichao Han12 Tingrui Zhang12

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: