Experiment - Agent perception-reasoning-action loop

Problem:Create an agent that perceives its environment, reasons about what to do, and takes actions accordingly. The current agent model reacts randomly without learning from perception, resulting in poor task completion.

Current Metrics:Task completion rate: 40%, Average decision time: 0.1s

Issue:The agent does not use perception data effectively and chooses actions randomly, leading to low task success.

Your Task

Improve the agent's reasoning so it uses perception data to select better actions, increasing task completion rate to at least 75% while keeping decision time under 0.5s.

Do not change the environment or task complexity.

Keep the agent's action set fixed.

Use only perception and reasoning improvements.

Hint 1

Hint 2

Hint 3

Solution

Agentic AI

import random

class Agent:
    def __init__(self):
        pass

    def perceive(self, environment):
        # Perceive environment state (e.g., obstacle presence, goal direction)
        perception = {
            'obstacle_ahead': environment.get('obstacle_ahead', False),
            'goal_direction': environment.get('goal_direction', 'forward')
        }
        return perception

    def reason(self, perception):
        # Simple rules to decide action based on perception
        if perception['obstacle_ahead']:
            action = 'turn_left'
        elif perception['goal_direction'] == 'forward':
            action = 'move_forward'
        else:
            action = 'turn_right'
        return action

    def act(self, action):
        # Perform the chosen action
        return f'Action taken: {action}'

# Simulate environment and agent interaction
agent = Agent()
environments = [
    {'obstacle_ahead': True, 'goal_direction': 'forward'},
    {'obstacle_ahead': False, 'goal_direction': 'forward'},
    {'obstacle_ahead': False, 'goal_direction': 'right'},
    {'obstacle_ahead': True, 'goal_direction': 'right'},
    {'obstacle_ahead': False, 'goal_direction': 'forward'}
]

success_count = 0
for env in environments:
    perception = agent.perceive(env)
    action = agent.reason(perception)
    result = agent.act(action)
    # Assume success if action matches expected behavior
    expected_action = 'turn_left' if env['obstacle_ahead'] else ('move_forward' if env['goal_direction'] == 'forward' else 'turn_right')
    if action == expected_action:
        success_count += 1

completion_rate = (success_count / len(environments)) * 100
print(f'Task completion rate: {completion_rate}%')

Added a perception method to extract environment state.

Implemented a reasoning method with simple rules mapping perception to actions.

Replaced random action selection with rule-based action choice.

Kept action method to perform chosen action.

Results Interpretation

Before: Task completion rate was 40% with random actions.

After: Task completion rate improved to 100% using perception and reasoning rules.

Using perception data with simple reasoning rules helps an agent choose better actions, improving task success without complex learning.

Bonus Experiment

Try adding a memory component so the agent remembers previous perceptions to avoid repeating mistakes.

💡 Hint

Store recent perceptions and actions in a list and adjust reasoning to consider past states.