Person detection using Hyper-redundant manipulator (snake robot) in inaccessible room using YOLO
Problem Statement:
In a critical, hazardous or damaged location, it's not a feasible solution to send a human for inspection of people or other things inside the building. In such situation, robots can help us a lot. The existing robots which does the search operations are quite big and hence they cannot carry out search operations through narrow spaces. For this a hyper-redundant robot like snake robot can be advantageous.
Goal:
-
To create a minimal design of snake robot in Solidworks and import the model and test world into gazebo for simulation and validation purpose.
-
To execute the linear progression motion of snake robot with a python gait code and tuning the configuration parameters for the simulation
-
To implement objects detection especially person detection in this case and to achieve the required detection accuracy.
Approach:
-
Designing a snake robot's module in Solidworks and assembling them for form a robot having 12 modules. Converting the assembly file to URDF in import into Gazebo world
​
​​
​
​​​
-
Configuring the controller yaml file according to the requirement. Creating the launch files and setting up the Gazebo world
-
Creating the python gait motion file to make the snake robot to move in linear regression and do the search operation. The general gait equation motion for snake robot is given by:
Pitch modules:
(pitch_offset) + (pitch_amplitude)*sin((temporal_frequency)*time + n*(spatial_frequency))
Yaw modules:
(yaw_offset) + (yaw_amplitude)*sin((temporal_frequency)*time + n*(spatial_frequency) + phase_difference)​
-
For the purpose, amplitude value of 0.65, spatial frequency of (2*pi)/6, temporal frequency of 1.5*pi and phase difference of 0 degrees were used
-
Implementing the YOLO package into the simulation and adjusting the parameters.
Results:
-
The designed module and hence the assembly of the snake robot is shown below
-
After modifying the yaml config file the robot is spawned to the environment along with a tunnel connecting to a inaccessible room with a person
-
After the snake robot entered the room, it has detected a person lying unconciously on the ground with 90% accuracy.
Media:​