Nonlinear Model Predictive Control (NMPC) for RoboCup SSL

This repository implements a Nonlinear Model Predictive Control (NMPC) framework for smooth trajectory planning and dynamic obstacle avoidance in the RoboCup Small Size League (SSL). It is designed as an alternative to reactive, kinematics-based approaches such as the Bang-Bang Trajectory Generator used in frameworks like TIGERs Mannheim’s Sumatra.

---

Overview

• Predictive obstacle avoidance with nonlinear constraints
• Optimization-based trajectory planning over a finite prediction horizon
• Cost function combines reference tracking, control effort minimization, and trajectory smoothness
• MATLAB-based implementation with solver-based optimization

---

Results

• Average velocity improvement of 5.3% over Bang-Bang control

  • MPC: 1.045 m/s
  • Bang-Bang: 0.992 m/s

• Generates smooth, efficient, and collision-free trajectories

• Maintains safety margins with predictive obstacle modeling

---

Methodology

1. System Model

   • State: position [x, y]
   • Control input: velocity vector [vx, vy]
   • Discrete-time integrator dynamics

2. Optimization Problem

   • Prediction horizon: N = 20 steps
   • Sampling time: Ts = 0.05 s
   • Cost function: reference tracking, control effort, smoothness

3. Constraints

   • Velocity limits: |vx|, |vy| ≤ vmax
   • Collision avoidance via nonlinear distance constraints with obstacle prediction

---

Future Work

• Multi-robot coordination for team play
• Integration with reinforcement learning for adaptive strategies
• Hardware deployment with real-time optimizations
• Sim-to-real validation on physical robots

---