基于轻量级安全通信协议的机器人快递交付系统

Research CV Security Robotics

项目概览

Overview

Design, analysis and implementation of a smart next generation secure shipping infrastructure using autonomous robot

GitHub github.com/Jiapei-Yang/Secure-Robotic-Shipping

论文 Paper doi.org/10.1016/j.comnet.2020.107779 · 开放稿open manuscript

面向机器人最后一公里配送：默认以轻量级密码学协议完成离线认证；扫码反复失败时切到视觉路径（行人检测 + Siamese Person Re-ID），找到收件人后再回到认证与交付。若超时仍未完成扫码、重识别多次失败或二次认证仍失败，则中止投递。整条链路在 TurtleBot3 真机上跑通，并用 ProVerif 对协议做了形式化分析。

Built for robotic last-mile delivery: the default path authenticates offline with a lightweight cryptographic protocol; after repeated QR failures it switches to vision (pedestrian detection + Siamese person re-identification), then returns to authentication and delivery once the recipient is found. If scanning times out, re-identification fails repeatedly, or secondary authentication still fails, the delivery is aborted. The full loop runs on a TurtleBot3, with the protocol analyzed in ProVerif.

一作发表 · Computer Networks (Elsevier), 2021 First author · Computer Networks (Elsevier), 2021 Yang, J., Gope, P., Cheng, Y., Sun, L. — Design, analysis and implementation of a smart next generation secure shipping infrastructure using autonomous robot. Yang, J., Gope, P., Cheng, Y., Sun, L. — Design, analysis and implementation of a smart next generation secure shipping infrastructure using autonomous robot.

我的角色：项目最初由三人分工——网络协议、计算机视觉、机器人平台。另外两人中途退出后，剩余工作由我独立接手：补学 ROS 并完成真机部署，设计与训练 Re-ID 模型，并把合作认证、非合作降级与安全分析接到同一条可演示链路。最终以一作将设计、实现与安全验证发表于同行评审期刊。 My role: the project started with three owners—network protocol, computer vision, and the robot platform. After the other two teammates left, I took over the remaining work: learning ROS and bringing up the hardware, designing and training the Re-ID model, and connecting cooperative auth, non-cooperative fallback, and security analysis into one demoable pipeline. I then published the design, implementation, and security validation as first author.

覆盖范围：合作式离线认证、非合作行人重识别、TurtleBot3 真机演示、ProVerif 安全分析
设计原则：交付必须可验证；认证可以降级，身份判断不能含糊 Scope: cooperative offline authentication, non-cooperative person re-ID, TurtleBot3 demo, ProVerif analysis
Design rule: delivery must be verifiable—fallback is allowed, ambiguous identity is not

核心能力

Capabilities

系统由三条互相咬合的能力组成：主路径认证、失败时的视觉降级，以及真机与安全验证。

Three interlocking pieces: the primary auth path, a vision fallback when that path fails, and hardware plus security validation.

合作式离线认证 Cooperative offline authentication 客户、服务端与机器人三端协议：一次性密钥、nonce 与 QR 密文比对。收件人出示二维码即可完成验证，无需长期在线。 A client–server–robot protocol with one-time keys, nonce, and encrypted QR matching. The recipient authenticates by presenting a QR code—no long-lived online session required.
非合作降级：检测 + Siamese Re-ID Non-cooperative fallback: detection + Siamese Re-ID 扫码连续失败后切换模式：机器人拍摄现场、检测行人，再用度量学习判断是否为客户本人；确认后再回到扫码交付。 After repeated QR failures, the robot captures the scene, detects pedestrians, and uses metric learning to decide whether the person matches the client—then returns to QR delivery.
真机闭环与形式化安全分析 Hardware loop and formal security analysis 在 TurtleBot3 上完成导航与交付演示；用 ProVerif 分析协议在泄露、伪造等威胁下的性质；并以一作论文固化设计与实验结论。 Navigation and delivery are demonstrated on TurtleBot3; ProVerif analyzes the protocol against threats such as disclosure and forgery; the design and results are recorded in a first-author paper.

系统流程

How it works

主路径走合作认证；扫码连续失败后进入视觉找人，成功后再回到扫码交付。超时未扫码、重识别多次失败，或二次认证仍失败时，任务中止。

Cooperative QR auth is the main path. After repeated scan failures, vision search runs; on success the system returns to QR delivery. Timeout without a scan, repeated Re-ID failures, or another failed auth round aborts the job.

1. 发起配送 1. Start delivery 客户向服务端请求配送任务 Client requests a delivery from the server

2. 签发凭证 2. Issue credentials 服务端生成 nonce 与一次性公私钥，分发给客户与机器人 Server generates a nonce and one-time key pair for the client and robot

3. 导航到交付点 3. Navigate to the drop-off 机器人按地图到达约定位置 Robot navigates to the agreed location

4. 合作认证（扫码） 4. Cooperative auth (QR) 扫描客户二维码，比对加密认证信息 Scan the client QR and compare encrypted auth payloads

首次扫码结果

First QR outcome

扫码成功

QR succeeds

完成交付 Complete delivery 身份确认，任务正常结束 Identity confirmed; job ends successfully

连续失败（如 5 次）→ 降级

Repeated failure (e.g. 5×) → fallback

5. 视觉找人（Re-ID） 5. Vision search (Re-ID) 拍摄现场 → 行人检测 → Siamese 比对客户图 Capture → detect pedestrians → Siamese match vs client image

视觉找人结果

Vision search outcome

确认为同一人

Same person confirmed

6. 切回合作认证 6. Return to cooperative auth 再次尝试扫码并比对 Retry QR scan and matching

二次扫码

Second QR attempt

成功

Success

完成交付 Complete delivery 匹配通过，任务结束 Match passes; job ends

再次失败 / 超时

Fail again / timeout

中止投递 Abort delivery 二次认证仍失败则停止任务 Stop the job if secondary auth still fails

多次识别失败 / 超时未确认

Repeated Re-ID failure / timeout

中止投递 Abort delivery 无法可靠确认收件人，任务停止 Cannot reliably confirm the recipient; job stops

另：到达后长时间未完成有效扫码（超时），同样进入「中止投递」，避免包裹在不确定身份下完成交接。

Also: if no valid scan completes within the timeout after arrival, the job aborts—so a parcel is never handed over under uncertain identity.

实现与演示

Demo evidence

下面是系统设计图与真机过程截图，点击可全屏查看。

System design figures and hardware captures below—click any image to enlarge.

系统总览

System overview

整体结构、双模式认证设计，以及实现拓扑。

Overall architecture, dual-mode authentication design, and implementation topology.

Loading
fig1

Loading
fig2

Loading
fig3

合作认证 · 真机

Cooperative authentication on hardware

机器人导航到达后扫描二维码，密文比对通过即完成交付。

After navigation, the robot scans the QR code; a ciphertext match completes delivery.

Loading
fig4

Loading
fig5

Loading
fig6

非合作降级 · Re-ID

Non-cooperative fallback with Re-ID

扫码失败后切换模式，检测行人并用重识别确认客户身份。

After QR failure, the system switches modes, detects pedestrians, and confirms identity with re-identification.

Loading
fig7

Loading
fig8

Loading
fig9

回到合作认证 · 完成交付

Return to cooperative auth and finish

找到收件人后再次扫码，匹配成功后结束任务。

Once the recipient is found, QR authentication runs again and the task completes on a successful match.

Loading
fig10

Loading
fig11

演示视频

Demo videos

真机与仿真录屏，托管在 YouTube。

Hardware and simulation recordings hosted on YouTube.

1. 合作认证真机

1. Cooperative authentication on hardware

youtube.com/watch?v=-cANuZxD9uQ

2. 非合作切换全流程

2. Full non-cooperative fallback flow

youtube.com/watch?v=lN5tngrdmds

3. Re-ID 仿真（Jupyter）

3. Re-ID simulation in Jupyter

youtube.com/watch?v=zgD6tC4vGLM

技术取舍

Tech choices

TurtleBot3 + ROS — 用真机验证导航与交付，而不只停留在仿真。 — to validate navigation and delivery on real hardware, not only in simulation.
Crypto + QR（hash / asymmetric / nonce） — 合作路径可离线验证，并减少长期密钥暴露。 — keeps the cooperative path offline-verifiable and limits long-lived key exposure.
Siamese Network · Person Re-ID — 用度量学习回答「是不是同一个人」，覆盖收件人不配合的场景。 — metric learning answers “same person or not” when the recipient cannot cooperate.
CUHK01 — 用于训练与 CMC 评测的公开行人重识别数据集。 — a public person Re-ID dataset for training and CMC evaluation.
ProVerif — 对合作协议做形式化分析，覆盖泄露、伪造等威胁模型。 — formal analysis of the cooperative protocol under threats such as disclosure and forgery.

结果与边界

Outcomes & boundaries

发表：一作发表于 Computer Networks（Elsevier, 2021），覆盖系统设计、真机实现与安全分析。
Publication: first-author paper in Computer Networks (Elsevier, 2021), covering system design, hardware implementation, and security analysis.
系统：合作认证与非合作 Re-ID 降级均可在真机路径演示；协议经 ProVerif 分析。
System: both cooperative auth and the non-cooperative Re-ID fallback are demoable on hardware; the protocol is analyzed with ProVerif.
边界：Re-ID 使用经典 Siamese 与 CUHK01，目标是配送系统里可靠的降级模块，而不是视觉方向的 SOTA 模型。
Boundary: Re-ID uses a classic Siamese setup on CUHK01—intended as a reliable fallback inside the delivery system, not a vision SOTA model.

论文开放稿： White Rose Research Online

Open-access manuscript: White Rose Research Online

项目概览

Overview

核心能力

Capabilities

系统流程

How it works

实现与演示

Demo evidence

系统总览

System overview

合作认证 · 真机

Cooperative authentication on hardware

非合作降级 · Re-ID

Non-cooperative fallback with Re-ID

回到合作认证 · 完成交付

Return to cooperative auth and finish

演示视频

Demo videos

1. 合作认证真机

1. Cooperative authentication on hardware

2. 非合作切换全流程

2. Full non-cooperative fallback flow

3. Re-ID 仿真（Jupyter）

3. Re-ID simulation in Jupyter

技术取舍

Tech choices

结果与边界

Outcomes & boundaries

项目链接

Links