PD Parameter Optimizer for Robotics Overview

This Python package optimizes Proportional-Derivative (PD) controller parameters for robotic systems by comparing policy output actions with actual executed actions. It uses sampling data from various PD parameter combinations to find optimal values that minimize tracking error.

Key features:

Loads and validates PD sampling data from CSV files

Calculates weighted tracking errors (position, velocity, acceleration)

Optimizes parameters using gradient-based methods

Performs performance analysis (settling time, overshoot, steady-state error)

Generates comprehensive visualizations

Saves optimization results in JSON format

Requirements

Python 3.7+
Required packages:
bash
numpy
pandas
matplotlib
scipy
scikit-learn

Installation

Clone the repository:
bash

git clone https://github.com/yourusername/pd-parameter-optimizer.git cd pd-parameter-optimizer

Install dependencies: bash

pip install -r requirements.txt

Usage Data Preparation

Create a folder for your sampling data (e.g., ./pd_sampling_data)

Save CSV files with the naming format: pd_p{P value}_d{D value}.csv

    Example: pd_p10.0_d0.5.csv

Each CSV file should contain these columns:

    timestamp

    policy_action

    actual_action

    position

    velocity (optional)

    acceleration (optional)

    target_velocity (optional)

    target_acceleration (optional)

Basic Example python

from pd_optimizer import PDParameterOptimizer

Configuration parameters (optional)

config = { 'file_pattern': '*.csv', 'p_range': (0.1, 50.0), 'd_range': (0.01, 5.0), 'weight_position': 1.0, 'weight_velocity': 0.5, 'weight_acceleration': 0.3 }

Initialize optimizer

optimizer = PDParameterOptimizer( data_folder='./pd_sampling_data', config=config )

Load data

optimizer.load_sampling_data()

Analyze existing PD parameters

analysis_df = optimizer.analyze_pd_performance() print("Top performing PD parameters:") print(analysis_df.head(5))

Optimize parameters

optimal_p, optimal_d, optimal_error = optimizer.optimize_pd_parameters() print(f"\nOptimal parameters: P={optimal_p:.4f}, D={optimal_d:.4f}") print(f"Minimum error: {optimal_error:.6f}")

Generate and save visualizations

optimizer.plot_results(save_path='optimization_results.png')

Save full results

optimizer.save_results('optimization_results.json')

Configuration Options Parameter Default Value Description file_pattern '*.csv' File pattern for data loading p_range (0.1, 200.0) Search range for P parameter d_range (0.01, 5.0) Search range for D parameter optimization_method 'L-BFGS-B' Optimization algorithm max_iterations 1000 Maximum optimization iterations tolerance 1e-6 Optimization tolerance weight_position 1.0 Weight for position error weight_velocity 0.5 Weight for velocity error weight_acceleration 0.3 Weight for acceleration error Output Visualization

The plot_results() method generates a 2x2 grid of visualizations:

Error Heatmap
Shows tracking error across different PD parameter combinations

Best Waveform Comparison
Compares policy actions vs actual actions for optimal PD parameters

PD Parameter Scatter Plot
Visualizes parameter space with error coloring and marks best point

Performance Metrics Comparison
Compares overshoot, settling time, and steady-state error for top 5 PD combinations

The analyze_pd_performance() method returns a DataFrame with these metrics for each PD combination: Metric Description Total_Error Weighted composite tracking error Settling_Time Time to reach within 2% of target value Overshoot Maximum overshoot percentage Steady_State_Error Average error in last 10% of trajectory Optimization Algorithm

The optimization process:

Uses sampled PD points as starting references

Estimates errors for unsampled points using distance-weighted interpolation

Applies gradient-based optimization (L-BFGS-B by default) to find minimum

Respects parameter boundaries defined in configuration

Contributing

Contributions are welcome! Please submit pull requests for:

Additional optimization algorithms
Improved interpolation methods
Enhanced visualization options
Extended performance metrics

License

This project is licensed under the MIT License - see the LICENSE file for details.