Risk-Aware Stochastic MPC for Run-Time Specifications in Signal Temporal Logic (RAM-RuTS)

Authors: Zengjie Zhang (z.zhang3@tue.nl) and Maico H.W. Engelaar (m.h.w.engelaar@tue.nl).

Model Predictive Control (MPC) for a stochastic linear system with runtime Signal Temporal Logic (STL) specifications.

Introduction

A practical engineering system is often required to accomplish given tasks with restricted risk levels. This can be achieved by risk-aware control incorporating the system's stochastic uncertainty, rendering a stochastic planning problem with signal temporal logic (STL) specifications.

Nevertheless, many practical applications need to dynamically assign tasks to the system during runtime. This requires the system to adjust or reschedule its current control strategy to accept the new task. Consider the following restaurant service robot scenario (as illustrated in Figure 1), where a service robot staying at the counter (HOME) is required to serve two tables (TARGETS) while avoiding collision with the wall (OBSTACLE). When short of power, it also needs to stay in the CHARGER for a certain period of time. These tasks may be assigned sequentially in the runtime, some of which are even subject to concurrency or conflict. An example of a task sequence may look like this:

• The robot starts from HOME at time $0$. It should not go outside the restaurant during the whole time and cannot collide with the wall;
• At time $5$, the robot receives a command "reaching either TARGET between time $20$ and $30$";
• At time $15$, the robot receives a command "reaching the CHARGER between time $20$ and $25$";
• At time $20$, the robot receives a command "visiting HOME between $25$ and $30$ and stay there for at least $5$ steps".

In this setting, the robot needs to reschedule its control strategy each time a new dynamic task is assigned. We expect the robot to accomplish as many dynamic tasks as possible. However, when the robot realizes its high risk of conflict or failure, it can reject the new task while fully focusing on the ongoing task. Figure 1 illustrates a solution where a robot accomplishes all assigned tasks. The blue line is its actual trajectory, implying sequential reaching "HOME $\rightarrow$ CHARGER $\rightarrow$ TARGET $\rightarrow$ HOME", while the lines in other colors are contemporarily planned trajectories that have been rescheduled in each stage.

Figure 1. A restaurant service robot scenario.

Approach

In a risk-aware control framework, we formulate the physical model of the robot as a single-integrator dynamic system with random dynamic noise and use a set of STL to describe the tasks above. In this sense, risk is quantified by the probability that a certain STL task is not satisfied. The risk also serves as evidence to support the decision-making of the robot whether a new task should be accepted or rejected. The control synthesis is performed using a tube-based shrinking-horizon model predictive control method. For the theoretical details, please refer to our associated ArXiv paper in https://arxiv.org/abs/2402.03165 which has been presented in the 8th IFAC Conference on Analysis and Design of Hybrid Systems, July 1-3, 2024.

Relation with Existing Toolbox

The probstlpy library in this project is modified from the stlpy toolbox.

Installation

System Requirements

Operating system

• Windows (compatible in general, succeed on 11)
• Linux (compatible in general, succeed on 20.04)
• MacOS (compatible in general, succeed on 13.4.1)

Python Environment

• Python version: test passed on python=3.11
• Recommended: IDE (VS code or Pycharm) and Conda
• Required Packages: numpy, treelib, matplotlib, scipy.

Required Libraries

• gurobipy solver (license required, see How to Get a Gurobi License)
• Python control toolbox (see Documentation)

Quick Installation

1. Install conda following this instruction;

2. Open the conda shell, and create an independent project environment;

conda create --name ram-ruts python=3.11

3. In the same shell, activate the created environment

conda activate ram-ruts

4. In the same shell, within the ram-ruts environment, install the dependencies one by one

conda install -c anaconda numpy
conda install -c conda-forge treelib
conda install -c conda-forge matplotlib
conda install -c anaconda scipy

5. In the same shell, within the ram-ruts environment, install the libraries

python -m pip install gurobipy
pip install control

6. Last but not least, activate the gurobi license (See How To). Note that this project is compatible with gurobi Released version 11.0.1. Keep your gurobi updated in case of incompatibility.

Running Instructions

• Run the main script main.py;
• Watch the terminal for runtime information;
• The figures will show up at the end of running; They are also automatically saved in the root directory;
• The figures may impede each other; Drag the figures for a better view;
• Check out the logging file INFO.log for the runtime information.

Fine-Tuning the Code

Feel free to try out the code with different parameter settings in the commons/configs.py file.

• Change the coordinates of the regions in the RG dictionary to construct a different map;
• Change the color map in the CM dictionary for preferred layout;
• Change the standard deviation variable Sigma for different noise levels;
• Change the initial position of the robot in x0;
• Customize the lists of runtime specifications specs and their instants in times for various tasks.

License

This project is with a BSD-3 license, refer to LICENSE for details.