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)
2. Navigation with Potential Fields

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:
- a deliberative one, containing a symbolic world model, which develops plans and
makes decisions in the way proposed by symbolic AI; and - a reactive one, which is capable of reacting to events without complex reasoning.