navigation and autonomous technologies

path planning

computing the most efficient route from a robot’s location to a specified destination

minimize risk and travel time

A*

an algorithm that finds the shortest path by estimating costs of moving from one point to another and adjusting its route in real time as obstacles are identified

RTT (rapidly-exploring random trees)

an algorithm that creates a map from paths known to be clear

obstacle avoidance

detect and circumnavigate barriers (users sensors like LIDAR)

sensor fusion

combining data from various sensors to help the robot accurately map its surroundings and make informed navigation decisions

adaptive algorithms

incorporation of machine learning into algorithms to learn from each navigation challenge encounters to help better avoid other obstacles

technology integration for efficiency

techniques such as keyframe selection in vSLAM reduce computational load by focusing on critical data, ensuring robots can respond swiftly to new info without being bogged down by processing delays

internal navigation systems (INS)

navigating without GPS

uses a combination of accelerometers and gyroscopes to estimate a robot’s position, orientation and velocity relative to a known starting point

accelerometeres

measures acceleration in robot’s frame, allowing for the calculation of velocity and displacement after integrating acceleration data over time

gyroscopes

detects changes in orientation by measuring rate of rotation around axes

crucial to correcting drift

dead reckoning

uses data from accelerometers and gyroscopes to continuously update estimated position and orientation

sensor fusion

combines INS data with other sensor inputs such as odometry or visual SLAM

periodic calibration

algorithms that recognize when a robot revisits a previously mapped location (loop closure) and allows for recalibration of INS

real time environmental mapping

SLAM technologies enable robots to create and update a map to navigate through dynamic environments

machine learning and ai for predictive navigation

robots can learn from past navigation experiences. over time, robots can predict potential hazards and adapt navigation strategies accordingly

increased efficiency and safety

dynamic path planning algorithms

allow the robot to recalculate its route on the fly when encountering unexpected obstacles

D* Lite

1. a robot has given goal coordinates in unknown terrain

2. Makes assumptions about the unknown part of the terrain (for example: that it contains no obstacles)

3. finds a shortest path from its current coordinates to the goal coordinates under these assumptions

4. robot then follows the path

5. observes new map information (such as previously unknown obstacles)

6. adds the information to its map and, if necessary, replans a new shortest path from its current coordinates to the given goal coordinates

7. repeats the process until it reaches the goal coordinates or determines that the goal coordinates cannot be reached

behavior based navigation

selects from repertoire of pre-defined behaviors based on current context

edge computing

the computer that does the algorithms is right there and has everything it needs downloaded already - no need to access the cloud

issues with automated adaptive navigation

needs communication with control center to update strategies based on human operators inputs or mission changes

computer vision

enables machines to interpret and understand the visual world through digital images and videos

interprets visual cues from surroundings allowing more precision

feature detection and tracking

can track specific landmarks, edges, corners or any distinct visual pattern

can identify and monitor the feature as it moves to ascertain location and orientation with respect to surroundings

depth perception and 3D mapping

stereo cameras and LIDAR together

object recognition and classification

detect objects and classify them through computer vision algorithms

traversable spaces vs obstacles, points of interest or locating survivors

semantic segmentation

partitions images into segments or pixels grouped by categories (road, debris, person, etc) and allows the robot to understand the environment in more detail with context

challenges and advances

challenges:

• varying light

• occlusions

• dynamic changes in scene

solutions:

• machine learning

• deep neural networks

terrain analysis

process including assessing various types of terrain to determine the most suitable navigation strategies

terrain classification

classified based on characteristics - flat, rocky, slippery or uneven

from a combination of cameras and LIDAR

machine learning important to process inputs

surface analysis

understanding texture, stability and incline to determine traction and risk of slipping or tipping over

uses photogrammy and 3D modelling

photogrammy

used in surface analysis to take measurements from photographs

obstacle detection and avoidance

uses computer vision and sensor fusion

adaptability

relies on sensory inputs, processing capabilities and locomotion mechanisms

challenges in computer vision

• interpretation of sensor data

• lighting conditions and weather