You are on page 1of 5

Asli_planning != AI.

planning
But marks = function(AI.planning, memory, skills.writing)

Bruhhh 😂
True wordssss :)

Sensorless planning
Sensorless planning, also known as conformant planning, is a method used in artificial
intelligence for planning actions in environments where the state of the world is not
fully known12. The goal of sensorless planning is to come up with a plan that works in all
possible cases.

In sensorless planning, the algorithm ensures that the plan should reach its goal at any
cost3. This is done by considering all possible states and actions, and coming up with
a plan that leads to the goal state regardless of the initial state and the effects of the
actions1.

For example, consider a robot in a grid world where the locations of obstacles are
unknown1. The robot needs to find a path to a goal location. In sensorless planning, the
robot would come up with a plan that guarantees it can reach the goal location from any
possible starting location, and regardless of where the obstacles are1.

However, sensorless planning can be computationally expensive and may result in overly
cautious plans, as it needs to consider all possible states and contingencies1. Therefore, it’s
typically used in situations where safety is critical, or where there’s no way to obtain more
information about the state of the world1.

Contingent planning
Contingent planning in artificial intelligence refers to planning under uncertainty, where
the outcomes of actions are not deterministic and can depend on external factors or
events. In contingent planning, the planner must consider multiple possible future scenarios
and develop a plan that is robust enough to handle a range of outcomes.

Let's illustrate contingent planning with an example involving a delivery robot:

### Example: Contingent Planning for a Delivery Robot

#### Initial State:


- **Robot Location:** Depot
- **Package Location:** Warehouse
- **Weather:** Unknown (either rainy or clear)

#### Goal State:


- **Robot delivers the package to the destination.**

#### Actions:
1. **Action:** Move to Warehouse
- **Preconditions:** Robot Location is Depot
- **Effects:** Robot Location becomes Warehouse

2. **Action:** Pick up Package


- **Preconditions:** Robot Location is Warehouse
- **Effects:** Robot is holding the package

3. **Action:** Move to Destination


- **Preconditions:** Robot is holding the package
- **Effects:** Robot Location becomes Destination

4. **Action:** Deliver Package


- **Preconditions:** Robot Location is Destination
- **Effects:** Package is delivered

#### Uncertain Events:


- **Weather:** May be rainy or clear, affecting the robot's movement reliability.

Now, let's create a contingent plan:

1. **Scenario A: Clear Weather**


- **Plan:**
1. Move to Warehouse
2. Pick up Package
3. Move to Destination
4. Deliver Package

2. **Scenario B: Rainy Weather**


- **Plan:**
1. Move to Warehouse (with additional precautions for rain)
2. Pick up Package
3. Move to Destination (with adjustments for rainy conditions)
4. Deliver Package

In this example, the planner has to account for the uncertainty of the weather. The plan
is contingent on the actual weather conditions encountered by the robot. If the
weather is clear, the robot follows the first plan; if it's rainy, it follows the second plan
with adjustments for the adverse conditions.

Contingent planning is essential in real-world scenarios where external factors can impact
the success of planned actions. It allows systems to be more adaptive and robust,
ensuring that plans can be executed successfully even in the presence of
uncertainties. Planning under uncertainty is a significant challenge in AI, and
contingent planning is one approach to address this challenge.
Online replanning
Online replanning in Artificial Intelligence (AI) is a process where an AI system revises its
plan of action based on new information or changes in the environment. This is particularly
important in dynamic environments where conditions can change rapidly¹.

One key point in online replanning is the choice of the initial state for the replanning process.
If the previous plan is still executing, determining the new plan's initial state can be
challenging.
One approach to this problem is to integrate the choice of the appropriate initial state into the
search process itself. The search is initialized with multiple candidate initial states, and a
time-aware evaluation function is used to prefer plans whose total goal achievement
time is minimal¹.

Another important aspect of online replanning is deciding when to replan. Replanning can
be triggered by specific events or changes in the environment. Events could include the
detection of an obstacle, the completion of a subtask, or the receipt of new information.
When an event occurs, the system initiates the replanning process to adapt to the new
circumstances.Replanning can be triggered when the sensor data indicates a significant
change in the surroundings. For example, if a surveillance drone detects a new obstacle,
it may initiate replanning to avoid the obstacle.

In summary, online replanning is a crucial aspect of AI systems operating in dynamic


environments. It involves making decisions about when to replan and what initial state to use
for the replanning process, among other factors. By effectively managing these aspects, AI
systems can adapt to changes and optimize their performance in real-world scenarios.

Example: Autonomous Vehicles:

In the context of autonomous vehicles, online replanning is crucial. A vehicle may encounter
unexpected road closures, traffic jams, or changes in weather conditions. The vehicle's
planning system needs to adapt its route and behavior in real-time to ensure safe and
efficient navigation.

Multi-agent planning:
Multi-agent planning involves coordinating the actions of multiple agents to achieve common
goals in a shared environment.
This field is particularly relevant in scenarios where multiple autonomous entities need to
work together to accomplish complex tasks.
Each agent in the system has its own objectives, capabilities, and knowledge, and the
challenge lies in orchestrating their actions to achieve a collective outcome.
When the environment consists of multiple agents, then the way a single agent plans its
action gets changed.
The environment can be co-operative or competitive. In both the cases agents actions
influences each other.
Multi agent planning strategies are as follows:-
Centralized Planning:
In centralized planning, there is a single, central authority responsible for generating plans
for all agents. This approach can lead to globally optimized solutions, but it may become
impractical as the number of agents increases or if the environment is dynamic.
This strategy is suitable for scenarios where there is a clear central decision-maker or a
global perspective is essential.

Decentralized Planning:
Decentralized planning distributes the planning process among the agents, and each agent
independently generates its own plan based on local information. There may be limited or no
communication between agents during the planning phase.
This strategy is useful when agents have partial knowledge of the environment or when
communication between agents is restricted or costly.

Cooperative Planning:
In cooperative planning, agents actively communicate and coordinate with each other during
the planning process. They share information about their plans, intentions, and the
environment to achieve a more efficient and coordinated outcome.
This strategy is effective in scenarios where collaboration among agents is beneficial, such
as multi-robot systems or collaborative decision-making.

—---------------------------------------------------- Uncertainty —------------------------------------------------

Rule based method for uncertain reasoning:


Rule-Based Reasoning, also known as rule-based systems or expert systems, is an
approach that uses a set of rules to make inferences or decisions. These rules are typically
in the form of "if-then" statements, and they serve as a knowledge representation
mechanism to guide reasoning processes. Rule-based systems are widely used in artificial
intelligence and expert systems to capture and apply expert knowledge in a specific domain.
Here's an elaboration on this approach with suitable examples:

Components of Rule-Based Reasoning:


Rule Base:
The rule base consists of a set of rules in the form of "if-then" statements. Each rule
encapsulates a piece of knowledge or a decision-making guideline.

Inference Engine:
The inference engine is responsible for applying the rules to the available data or evidence
to derive conclusions. It determines which rules to fire based on the current state of the
system.

Fact Base:
The fact base contains the current state or observations of the system. It represents the
input data that the inference engine uses to match against the conditions specified in the
rules.
Example:
Let's consider a simple example of a rule-based system for diagnosing a medical condition,
such as a fever. The rule base consists of rules written in the form "if-then" statements:

Rule Base:

Rule 1: If body temperature is higher than 100 degrees Fahrenheit


Then diagnose fever.

Rule 2: If patient reports muscle aches and fatigue


Then diagnose flu.

Rule 3: If fever persists for more than three days


Then recommend a visit to a doctor.

Fact Base (Current Observations):


Fact 1: Body temperature = 101 degrees Fahrenheit.
Fact 2: Patient reports muscle aches and fatigue.

Inference Engine:
The inference engine evaluates the rules based on the provided facts.
Rule 1 is fired because the body temperature is higher than 100 degrees Fahrenheit.
Rule 2 is fired because the patient reports muscle aches and fatigue.
The system concludes that the patient may have both a fever and the flu.

Result:
The system provides a diagnostic recommendation based on the fired rules.

You might also like