Group Surfing: A Pedestrian-based Approach to Sidewalk Robot Navigation
這篇論文主要研究如何在人行道上給機(jī)器人導(dǎo)航诚镰。論文根據(jù)人行橫道上行人的多與少，來使用兩種不同的算法為機(jī)器人導(dǎo)航。在行人密集(pedestrian-rich)的環(huán)境下，使用Group Surfing方法，模仿前方的行人肯适，以遵守規(guī)則(socially-compliant)的行為來避開行人和障礙物墅诡，并最后抵達(dá)目的地望迎；在行人稀疏(pedestrian-sparse)的環(huán)境下障癌，通過檢測(cè)馬路沿，沿著馬路沿導(dǎo)航擂煞。在這兩種算法下混弥，底層的避障模塊是有模仿人類傾向(human-aware)。最后对省，作者在仿真和實(shí)物上均驗(yàn)證了算法蝗拿。
總體而言，這篇文章跟蹤行人的思路比較好蒿涎，算法實(shí)現(xiàn)細(xì)節(jié)清楚哀托，在算法驗(yàn)證階段有條理，總體質(zhì)量比較高劳秋。
多平臺(tái)維護(hù)不易仓手，內(nèi)容實(shí)時(shí)更新于 個(gè)人網(wǎng)站，請(qǐng)移步閱讀最新內(nèi)容玻淑。
INTRODUCTION
本文的一個(gè)側(cè)重點(diǎn)是嗽冒，人機(jī)交互，強(qiáng)調(diào)機(jī)器人盡量不影響行人的行為补履，或者以人的方式去交互添坊。所以，研究目標(biāo)不是一個(gè)簡(jiǎn)單的(類似于最短路徑)求最優(yōu)解的問題箫锤。
首先贬蛙，這篇文章的應(yīng)用場(chǎng)景是人行道(Sidewalk)，導(dǎo)航空間在路旁谚攒，行人的方向大概是兩個(gè)線性的阳准。
Sidewalks present a unique yet challenging environment in that the navigable space combines elements of both roads and free indoor spaces. Often sidewalk motion is restricted to two linear directions and the resulting navigable space is limited, like on roads.
問題的復(fù)雜性在于，行人的活動(dòng)比較隨機(jī)馏臭，可能聚團(tuán)(group)一起走野蝇。
However, pedestrians generally do not walk in perfect queues. Instead, people tend to walk in groups of variable sizes and speeds and move along with a general self-organizing crowd flow.
人行道導(dǎo)航(sidewalk navigation)的特點(diǎn)是，必須考慮行人的隨機(jī)行為括儒，同時(shí)遵守一定的社會(huì)規(guī)則浪耘，比如行人的交互常識(shí)(pedestrian conventions)，人際距離學(xué)(appropriate proxemics)塑崖，即行走方向保持的距離比垂直方向的距離要遠(yuǎn)。
Compared to autonomous road navigation, sidewalk navigation must also account for stochastic human movement that necessitates dynamic obstacle avoidance. Furthermore, certain social rules, such as walking in lanes or affording more space in the direction of walking than in the perpendicular direction, are rules that a robot should follow as well.
已有的研究方法有推理(reasoning)和學(xué)習(xí)(learning)痛倚，或者兩者的結(jié)合规婆。社會(huì)規(guī)則(Social rules)，在推理方法中作為代價(jià)函數(shù)，或在學(xué)習(xí)方法中作為獎(jiǎng)勵(lì)函數(shù)抒蚜。
In general, methods are based on either reasoning, learning, or a combination thereof.
已有方法的局限性在于掘鄙，利用對(duì)行人意圖的理解，通常適用于開放環(huán)境(open environments)嗡髓，而本文的研究問題是有限制的人行道環(huán)境操漠，同時(shí)沒有考慮人行橫道的物理邊界，機(jī)器人的運(yùn)動(dòng)會(huì)影響行人流等饿这。
Here, the aforementioned approaches may be less effective as they do not account for the physical sidewalk boundaries, or how robot movement will affect pedestrian flow.
所以浊伙，本文的主要研究問題是如何考慮附近行人的行為和行人流，機(jī)器人最終抵達(dá)終點(diǎn)长捧。
The key research question this paper considers is how mobile robots can utilize nearby pedestrian behaviours and flows to navigate towards a global goal.
針對(duì)這個(gè)問題嚣鄙，作者給出的答案是，模仿行人的行為串结。當(dāng)檢測(cè)到有行人朝向機(jī)器人的終點(diǎn)時(shí)哑子，會(huì)允許機(jī)器人模仿并從用行人的社交行為。
When our navigation stack detects people moving towards the robot’s goal, a ‘group surfing’ behaviour is used. This allows the robot to imitate and participate in pedestrian social behaviours.
在行人稀疏或者簡(jiǎn)單的人行道肌割，距離人行道的路沿一段距離卧蜓，沿著這個(gè)向目標(biāo)走(trajectory following)。
In an unpopulated and simple sidewalk environment, the default behaviour is to follow a trajectory offset from the sidewalk curb towards the goal.
SYSTEM
系統(tǒng)的設(shè)計(jì)比較中規(guī)中矩把敞，有新意的是弥奸，在航路點(diǎn)外，生成動(dòng)態(tài)的子目標(biāo)點(diǎn)(subgoals)先巴。
- 首先利用Google Maps’ API去生成高層次的航路點(diǎn)(waypoint)其爵；
- 導(dǎo)航模塊(sidewalk navigation module)根據(jù)附近是否有行人流(nearby pedestrian flow)來采用group surfing或者sidewalk following方法。
- 無論哪種方法伸蚯，對(duì)外輸出是子目標(biāo)點(diǎn)(subgoals)摩渺，這是避障算法的輸入，最后輸出速度矢量(velocity command)指令剂邮，控制機(jī)器人的運(yùn)動(dòng)摇幻。
- 周期輪訓(xùn)，是否抵達(dá)當(dāng)前目標(biāo)點(diǎn)挥萌。
METHODS
Group Surfing
核心思路和目標(biāo)：模仿人類的自然行為绰姻，包括沿路行走(walking in lanes)，避障(avoiding collisions with other pedestrians or obstacles)引瀑，路口等待(waiting at intersections to cross)狂芋，不走入交通中(not walking into traffic)。類似于仿生學(xué)憨栽，這次模仿的只不過是人類自己罷了帜矾。
- Filter Candidate Groups
- 過濾那些遠(yuǎn)離航路點(diǎn)的行人組翼虫。
Filter out groups moving away from the waypoint.
- 重點(diǎn)解釋下，
的意義屡萤。在我看來珍剑，這個(gè)并沒有物理意義，作者只不過想利用向量點(diǎn)乘的正負(fù)來判斷行人是否在遠(yuǎn)離航路點(diǎn)死陆。ps. 向量點(diǎn)乘的正負(fù)性取決于余弦角招拙，只要兩個(gè)向量的夾角小于90度(向目標(biāo)點(diǎn)靠近)，就是正值措译。
If this value is non-positive, discard Gi as a subgoal candidate.
- Smart Group Selection
- 方法核心思路和目標(biāo)：從篩選后的行人組中别凤，選擇平均速度小于且最接近于機(jī)器人最大速度的作為最優(yōu)跟蹤組(the optimal group to follow)。選擇該組中距離機(jī)器人最近行人的當(dāng)前位置作為子目標(biāo)點(diǎn)(subgoals)瞳遍。
Once we have filtered out unsuitable groups, the algorithm selects the optimal group to follow.
- 那么闻妓，路徑規(guī)劃和避障的問題就是在機(jī)器人當(dāng)前位置和最優(yōu)跟蹤組的最近行人之間，規(guī)劃出一條無碰撞的路徑掠械。
We intentionally select the closest person as a subgoal as attempting to reach the average group position could lead to path planning through pedestrians located between the average group position and the robot’s current position.
Curb Following
方法核心思路和目標(biāo)：使用3D laser sensor采集點(diǎn)云由缆，然后利用Random Sample Consensus (RANSAC)算法去識(shí)別馬路沿。
We make use of contextual knowledge; sidewalks are normally surrounded by streets and buildings or empty space. Our robot first acquires a surrounding point cloud using a 3D laser sensor and filters out points that are at the same height as or above the plane defined by the robot wheel contacts.
Collision Avoidance
- Human-Aware Collision Avoidance
- 在group surfing 和 curb following中猾蒂，使用已有的學(xué)習(xí)方法均唉，Socially-Aware Collision Avoidance with Deep Reinforcement Learning (SA-CADRL) 來作為避障算法。
Socially-Aware Collision Avoidance with Deep Reinforcement Learning (SA-CADRL), as the collision avoidance component of our navigation stack. The collision avoidance system navigates to a local subgoal generated by either the group surfing or the curb following approach.
- 其中肚菠，引入社交獎(jiǎng)勵(lì)函數(shù)來鼓勵(lì)社會(huì)行為舔箭。
The reinforcement training process induces social awareness through social reward functions, which give higher values to actions that follow social rules.
- Static Obstacle Avoidance:
- 把靜態(tài)障礙物作為靜態(tài)的行人，仍然使用SA-CADRL算法來處理蚊逢。
We also use SA-CADRL to avoid these static obstacles by adding “static pedestrians” to the state vector.
SIMULATION DEMONSTRATION AND EXPERIMENTS For
仿真環(huán)境的構(gòu)建：使用ROS和Gazebo仿真套件层扶。
We use the Robot Operating System (ROS) and Gazebo simulator suite. To simulate pedestrians, we use the Pedsim ROS library, which relies on the social force model.
Simulation Demonstration
在虛擬環(huán)境中，以GPS校準(zhǔn)烙荷，重新搭建了周圍的環(huán)境镜会，復(fù)制了機(jī)器人模型。
Simulation Experiments and Evaluation
目的：驗(yàn)證機(jī)器人可以沿著社會(huì)接受(socially-acceptable)的行為终抽，最后抵達(dá)終點(diǎn)戳表，即機(jī)器人的路徑和行人的路徑是相似的。
In evaluating our navigation system, our main goal was
to show that the system successfully navigates the robot to its final goal through a socially-acceptable path. That is, the path that our robot takes to the goal is similar to what a pedestrian would take to the same goal.
首先昼伴，在虛擬環(huán)境中匾旭，使用提出的算法抵達(dá)終點(diǎn)；然后圃郊，文中引入對(duì)比試驗(yàn)价涝，使用最短路徑的方法抵達(dá)終點(diǎn)。
We tracked the path taken by the robot and the path taken by a simulated pedestrian . We also tracked the shortest path that the robot could take within the confines of the sidewalk.
現(xiàn)在持舆，有三組軌跡飒泻，行人的真實(shí)軌跡鞭光，現(xiàn)有算法軌跡，最短路徑的軌跡泞遗。本文使用an independent samples t-test等數(shù)學(xué)方法，比較兩個(gè)軌跡和人類實(shí)際軌跡的相似度席覆，來證明現(xiàn)有的算法更加符合人類的行為史辙。贊一下，比較的有理有據(jù)佩伤。
HARDWARE DEMONSTRATION
Hardware Setup
機(jī)器人的配置如下：
We use the PowerBot from Omron Adept Mobile Robots as our differential drive mobile base. The robot is equipped with multiple sensors: a Velodyne VLP-16 3D LiDAR sensor; a SICK LMS-200 2D laser sensor; a RealSense RGB- D sensor, and GPS and IMU sensors. Our PowerBot’s max speed is 0.8m/s. This limits its capacity of following faster pedestrian groups.
Demonstration and Discussion
- 使用SPENCER算法來識(shí)別行人和行人流聊倔。
- 原打算使用lidar來識(shí)別行人的腿部，但是錯(cuò)誤率太多生巡，使用SPENCER提供的RGB-D-based的上身識(shí)別算法耙蔑。
- 對(duì)機(jī)器人局限的探討，以及算法的不足之處孤荣，從工程的角度討論甸陌。
CONCLUSIONS
待提高的地方：
- For the group surfing component, one main area for improvement is in the selection process of groups to imitate.
- Criteria: group velocity; group trajectory; group size
- External observers of the group surfing behaviour will be interviewed to gauge if the imitation behaviour is socially acceptable.
- For collision avoidance, a more specialized technique would allow for more efficient navigation.
- We hope to decouple static collision avoidance from dynamic collision avoidance.
- For curb following, our approach only works for sidewalks that limit directly to the street, ignoring common tree belt, median, hellstrip, etc. Our future plan is to introduce detection and recognition of these non-transitable areas and incorporate them in our navigation module.
