【COMP329 LEC 2 Agent and Robot Architectures】

Agent and Robot Architectures

Part 3 Reactive Architectures and the Subsumption Architecture

In these lectures, we look at alternative architectures that better support some classes of agents and robots
• At the end, we then examine how hybrid architectures exploits the best aspects
of deliberative and reactive ones

1. Agent Control Loop as Layers

连续地进行

2. Behaviour based control

并排的进行??

在最后整合到Coordination的过程中,有多种选择的方式,包括:

a. Pick the ``best''

b. Sum the outputs
c. Use a weighted sum

3. Brooks Behavioural Languages

4. Emergent Behaviour

我理解的就是,当一些规则同时进行的时候,可能会产生出一些反应,使得行为变成"程序以外"的结果,比如第二张图片.

synergies:协同,配合。(这里指的是 轻微偏右运动 + 躲避障碍 = 墙体跟随 这个协同)


5. Subsumption Architecture

就是有很多行为指令,但每个行为指令都有优先等级,例如"躲避障碍"就是一个底层指令,"more primitive kinds of behaviour",更原始行为。

层层分级使得他特别强大

复杂行为由简单行为组成

每个行为都独立,所以能被独立地:被编码 / 检测 / debugged

Higher level behaviours inhibit(抑制) lower levels

Part 4 Subsumption Architecture Examples

1. Steel's Mars Explorer System

2. ToTo

3. Summary

Part 5 Potential Fields and Hybrid architectures

1. Potential Fields 人工势能场


1. Simple fields can be combined to model complex environments

(a. Uniform - guides the robot in a straight line (useful for following a corridor)

(b. Perpendicular - pushes the robot away from linear obstacles( good for modelling large obstacles or walls)

(c. Tangental - guides the robot around an obstacle

(d. Attractive - draws the robot to a point (useful for defining weigh points in a path)

(e. Repulsive - pushes the robot away a point (good for modelling obstacles)

3. Potential Fields 的优缺点
1. Advantages

Easy to visualise
Easy to combine different fields

2. Disadvantages

High update rates necessary
Parameter tuning is important

2. Hybrid Architectures

To build a agents, neithor a completely deliberative nor completely reactive approach is suitable
An obvious approach is to build an agent out of two (or more)
subsystems:

  1. a deliberative one, containing a symbolic world model, which develops plans and
    makes decisions in the way proposed by symbolic AI; and
  2. a reactive one, which is capable of reacting to events without complex reasoning.
相关推荐
微功夫信息技术5 小时前
分层多智能体强化学习驱动的非急救转运公平 - 效率统一调度系统研究与实践
人工智能·学习·算法·动态规划
dadaobusi6 小时前
学习:开源项目Cheshire
学习
撩得Android一次心动6 小时前
Jetpack Compose 知识点整理1【个人用】
android·学习·kotlin·android jetpack·compose
xian_wwq8 小时前
【学习笔记】OWASP Agentic 应用安全(二)
笔记·学习·安全·owasp
小弥儿9 小时前
GitHub今日热榜 | 2026-09-04:Agent省 token 成今日主线
学习·开源·github
前端精髓9 小时前
NestJS 是什么(对着 Spring Boot 一起学习)
spring boot·后端·学习
kyrie_sakura10 小时前
MySQL数据库学习笔记2--系统函数(分组,单行,窗口函数)
数据库·学习·mysql
dadaobusi10 小时前
学习:XS-Gem5参数
学习
LearnYard11 小时前
技术博主实测:2026年大语言模型辅助学习工具横向对比
人工智能·学习·语言模型
M78佐菲11 小时前
HTML学习笔记
linux·笔记·学习·tcp/ip·html