《電子技術(shù)應(yīng)用》
您所在的位置:首頁(yè) > 其他 > 設(shè)計(jì)應(yīng)用 > 基于深度強(qiáng)化學(xué)習(xí)和社會(huì)力模型的移動(dòng)機(jī)器人自主避障
基于深度強(qiáng)化學(xué)習(xí)和社會(huì)力模型的移動(dòng)機(jī)器人自主避障
網(wǎng)絡(luò)安全與數(shù)據(jù)治理 2023年3期
李恒,劉輕塵,馬麒超
(中國(guó)科學(xué)技術(shù)大學(xué)信息科學(xué)技術(shù)學(xué)院,安徽合肥230026)
摘要: 深度強(qiáng)化學(xué)習(xí)在移動(dòng)機(jī)器人自主避障領(lǐng)域已得到廣泛應(yīng)用,其基本原理是通過(guò)模擬環(huán)境中的不斷試錯(cuò),結(jié)合獎(jiǎng)勵(lì)機(jī)制提升機(jī)器人的避障性能。然而,針對(duì)不同任務(wù)場(chǎng)景,網(wǎng)絡(luò)訓(xùn)練效率存在顯著差異。同時(shí),在人群密集的場(chǎng)景中,機(jī)器人的行為可能對(duì)人類造成干擾。為了應(yīng)對(duì)訓(xùn)練效率低下和機(jī)器人行為不符合社會(huì)規(guī)范的問(wèn)題,提出了一種將社會(huì)力模型融入深度強(qiáng)化學(xué)習(xí)的自主避障策略。該策略首先將人類未來(lái)的運(yùn)動(dòng)軌跡考慮進(jìn)獎(jiǎng)勵(lì)函數(shù),以確保機(jī)器人理解人類意圖并避免闖入人類的舒適區(qū)。其次,在訓(xùn)練過(guò)程中引入先驗(yàn)的傳統(tǒng)控制器模型,并設(shè)計(jì)了一種基于概率的切換開(kāi)關(guān),以隨機(jī)切換控制器輸出,提高機(jī)器人的探索效率。實(shí)驗(yàn)結(jié)果表明,所提出的方法能夠增加機(jī)器人與人類之間的安全距離,同時(shí)實(shí)現(xiàn)平穩(wěn)導(dǎo)航。
中圖分類號(hào):TP273
文獻(xiàn)標(biāo)識(shí)碼:A
DOI:10.19358/j.issn.2097-1788.2023.03.011
引用格式:李恒,劉輕塵,馬麒超.基于深度強(qiáng)化學(xué)習(xí)和社會(huì)力模型的移動(dòng)機(jī)器人自主避障[J].網(wǎng)絡(luò)安全與數(shù)據(jù)治理,2023,42(3):68-73,79.
Autonomous obstacle avoidance for mobile robots based on deep reinforcement learning and social force model
Li Heng,Liu Qinchen,Ma Qichao
(School of Information Science and Technology, University of Science and Technology of China, Hefei 230026, China)
Abstract: Deep reinforcement learning has been widely applied in the field of mobile robot autonomous obstacle avoidance Its basic principle is to simulate continuous trialanderror in the environment and improve the robot’s obstacle avoidance performance by combining reward mechanisms However, the training efficiency of the network varies significantly depending on the task scene, and in crowded scenes, the robot’s behavior may cause interference with humans To address the problems of low training efficiency and robots behaving inappropriately, this paper proposes a selfobstacle avoidance strategy that incorporates the social force model into deep reinforcement learning The strategy firstly considers the future trajectory of humans in the reward function to ensure that the robot understands human intentions and avoids entering the human comfort zone Secondly, during the training process, a priori traditional controller model is introduced and a probabilitybased switching method is designed to randomly switch controller outputs to improve the robot’s exploration efficiency The experimental results show that the proposed method can increase the safety distance between the robot and humans while achieving smooth navigation.
Key words : eep reinforcement learning; social force model; autonomous obstacle avoidance

0    引言

自主避障是移動(dòng)機(jī)器人應(yīng)用中的基礎(chǔ)技術(shù),其可以確保機(jī)器人在機(jī)場(chǎng)和購(gòu)物中心等人流擁擠場(chǎng)景中實(shí)現(xiàn)安全導(dǎo)航。人類有觀察他人以調(diào)整自身行為的能力,因此可以輕松穿過(guò)人群。然而,在高度動(dòng)態(tài)和擁擠的場(chǎng)景中進(jìn)行自主避障仍然是移動(dòng)機(jī)器人的一項(xiàng)艱巨任務(wù)。傳統(tǒng)導(dǎo)航框架中的避碰模塊通常將動(dòng)態(tài)障礙物視為靜態(tài),例如動(dòng)態(tài)窗口方法(DWA),或者僅根據(jù)某些交互規(guī)則關(guān)注下一步行動(dòng),例如互惠速度障礙(RVO)和最優(yōu)互惠碰撞避免(ORCA)。由于這些方法僅通過(guò)被動(dòng)反應(yīng)防止碰撞,并且通常使用人為定義的函數(shù)以保證安全,因此會(huì)導(dǎo)致機(jī)器人的運(yùn)動(dòng)不自然、短視和不安全。相比之下,強(qiáng)化學(xué)習(xí)導(dǎo)航技術(shù)可以通過(guò)不斷地探索和學(xué)習(xí)增強(qiáng)機(jī)器人的感知能力,從而實(shí)現(xiàn)更有力的決策。




本文詳細(xì)內(nèi)容請(qǐng)下載:http://theprogrammingfactory.com/resource/share/2000005258




作者信息:

李恒,劉輕塵,馬麒超

(中國(guó)科學(xué)技術(shù)大學(xué)信息科學(xué)技術(shù)學(xué)院,安徽合肥230026)


微信圖片_20210517164139.jpg

此內(nèi)容為AET網(wǎng)站原創(chuàng),未經(jīng)授權(quán)禁止轉(zhuǎn)載。