The motivation behind this project is the rehabilitation of motor skills. In this case, the inverted pendulum is used as a game, where the user moves the cart left or right to balance the pendulum.
To reduce frustration and tailor difficulty for different users, Model Predictive Control (MPC) is used to compute optimal control inputs. These control inputs are then fed back to the user through haptics.
Alternative simpler methods were also tested, such as applying force output based on the relative angle away from vertical. However, this only guided the user in the general direction without conveying the intuition of other states. By utilizing MPC, the cost function can be fine-tuned to consider not only the pendulum's angle but also the cart's position, leading to more effective assistance.
This implementation requires Linux and assumes that acados is already set up. For installation details on acados, refer to their GitHub page.
python3 -m venv inverted_venv
source inverted_venv/bin/activate
pip install -r requirements.txtIn separate terminals, run the following commands:
python3 main.pypython3 Haptics.pyThe simulation will work without a haptic device, but there will be no force feedback.
- Press 'E' to start the game
- Press 'M' to toggle MPC on/off
- Press 'R' to restart after game over
- Press 'Q' to quit the game
- The pendulum simulation follows realistic physics and considers torque, gravity, and damping effects.
- MPC computes optimal control inputs in real time to help the user maintain balance.
- A haptic device (if available) provides feedback force to guide the user.
- Scoring is based on the angle of the pendulum and accumulated over time.
