【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
石像鬼₧魂石5 小时前
HexStrike-AI人工智能 渗透测试学习(Metasploitable2 192.168.1.4)完整流程总结
学习·ubuntu
非凡ghost5 小时前
MusicPlayer2(本地音乐播放器)
前端·windows·学习·软件需求
蓝桉~MLGT6 小时前
Ai-Agent学习历程—— 阶段1——环境的选择、Pydantic基座、Jupyter Notebook的使用
人工智能·学习·jupyter
杰米不放弃7 小时前
AI大模型应用开发学习-26【20251227】
人工智能·学习
执笔论英雄7 小时前
【RL】Megatron使学习forward_backward_func返回值
学习
幺零九零零8 小时前
压测学习-JMeter
学习·jmeter
程途拾光1589 小时前
自监督学习在无标签数据中的潜力释放
人工智能·学习
软件技术NINI9 小时前
JavaScript性能优化实战指南
前端·css·学习·html
Blossom.1189 小时前
多模态大模型LoRA微调实战:从零构建企业级图文检索系统
人工智能·python·深度学习·学习·react.js·django·transformer
一 乐10 小时前
健身房预约|基于springboot + vue健身房预约小程序系统(源码+数据库+文档)
java·数据库·vue.js·spring boot·后端·学习·小程序