---
title: Overview on Robotics
tags: cs 593 robotics
---

# Overview on Robotics

-> cognition - >
| |
-< body <----

perception -> planning -> control -> environment loops back to perception

**draw that out lol that sucked**

## perception
- local perception
- camera on the robot
- e.g. depth sensor
- global perception
- assume we know the entire space
- usually generated through SLAM
- simultaneous localization and mapping
- pre-built packages (omni mapper)
- continuing research with SLAM
- data association
- model-based vs model-free
- landmark detection
- modality
- touch
- thermal
- vision
- combining modalities

## planning
- high-level planning
- task planning
- discrete spaces
- low-level planning
- deals with the environment and actuators
- for each task, how will the robot send commands to its actuators to interact with the environment
- continuous spaces
- tools
- discretization-based approaches
- A^*^
- disadvantages
- curse of dimensionality
- resolution is important
- sampling based methods
- RRT, RRT*, PRMs
- advanced constraints
- constraints
## control / action
- send comands to actuators to follow waypoints to get to targets
- Kalman filters to incorporate system noise (uncertainty) into planning algorithm

### definitions
- localization
- knowing where i am with reference to another object
- configuration space (c-space)
- robot joint space
- work/task space
- can be 2D or 3D
- point mass
- configuration space
- (x,y) -> point -> R^2^
- workspace
- 2D -> R^2^
- rigid-body
- configuration space
- (x,y, $\theta$ ) -> point with angle -> R^3^
- can get to state-space
- workspace
- 2D -> R^2^
- planning algorithms
- plan in configuration space

### course specifics
- assume global perception is provided or assume local perception
- emphasis on low-level planning and assume high-level plan is given
- combine machine learning with sampling based planning
- focus on sampling based planning

### course logistics
https://qureshiahmed.github.io/sp22.html
- assignment 1
- sampling based methods
- assignment 2
- PID for robot arm to follow circular trajectory
- assignment 3
- combining learning with planning
- assignment 4
- deep reinforcement learning
- final project
- use techniques not covered in assignments!
- class participation
- answering questions on piazza

### read up on before next class

- Modern robotics book: http://hades.mech.northwestern.edu/images/7/7f/MR.pdf

configuration space and workspace and state-space !!!!

RRT (rapidly-exploring random tree)
-
RRT*, PRMs

A*
- graph traversal and path search algorithm
- combines dijkstra's algorithm and greedy best-first-search
- dijkstra's
- favoring vertices that are close to the starting point
- greedy best-first-search
- favoring vertices that are close to the goal

PID controller(proportional-integral-derivative controller)
- single input, single ouput
- control loop mechanism that modifies itself based off of derivatives and proportions to minimize error
- e.g. cruise control on a car, power requirement changes if you go up a hill but speed doesn't change

MPC (model predictive controller)
- multiple inputs, multiple outputs
- multivariable control algorithm that uses
- an internal dynamic model of the process
- a cost function over the receding horizon
- an optimization algorithm minimizing the cost function using the control input

think of example where workspace has higher dimension than c-space