You are on page 1of 1

Sample New and Complex function idea:

Development of coupled Longitudinal and Lateral Vehicle Modeling of ActiveShuttle to analyse its Object
and Event Detection and Response on uneven/ inclined surfaces.
Why it is new and complex?
We know that the current prominent challenges in AV industry are Navigation through crowded urban areas,
Heavy rain/ fog, and rolling hills. Assuming that our Activeshuttle would not be exposed to the first two cases,
I chose this complex function of analysing Activeshuttle's performance on uneven terrains to detect objects
and it is new because of its novel design and non-uniform load parameters in the intralogistics facility.
Methodological approach accordance to selected topic:
1. Determination of software and hardware architecture:
Hardware: selection of computing hardware and exteroceptive/ proprioceptive sensors in consideration of all
desired technical specifications such as FOV, resolution, dynamic range, and so on.
Software: Modeling of full-stack software architecture which includes key areas such as Environment
perception and Mapping, Motion planning, Vehicle control, and System supervision system.
2. Longitudinal, lateral, and Tire Modeling for vehicle's Kinematics and Dynamics:
Longitudinal Modeling: Using PID controller which integrates feedforward and feedback control.
Lateral Modeling: Using pure pursuit/ Stanley Geometric controller and Model predictive Control.
Tire Modeling: Using well known parameter based Pacejka Tire model (covers both lateral and longitudinal).
3. Methods for Object and Event Detection and Response (OEDR).
State estimation/ Localization: Using recursive least squares and Maximum likelihood estimator algorithm
combined with unbiased and consistent Kalman filters to fuse multi-sensor data for single state estimation.
Visual perception:
▪ Calibration of Stereo Cameras to project 3D points from the world into the camera imaging sensor.
▪ Using stereo cameras and modeling it to estimate depth since scale info gets lost as 3D points are
projected as 2D on camera image plane. This requires the use of specialized disparity computation
algorithms.
▪ Considering parallel optical axes, Constrain the correspondence search to be along the epipolar line,
reducing the search from 2D to 1D.
▪ Considering non-parallel optical axes, we can perform stereo rectification using computer vision
packages such as MATLAB and OpenCV.
▪ Image Filtering through cross corelation and convolution operations.
▪ Apply feature detection, description and Matching algorithms for localization and Mapping.
▪ Develop and train Convolutional Neural networks for 2D object detection and sematic segmentation.
In addition, I also have good knowledge on the most important Motion planning using Mission planner,
Behaviour planner and Local planner to generate planned trajectory for vehicle control. Finally, I know about
system supervision system architecture and possess advanced knowledge on safety levels in AV.
Virtual simulation system to be used: CarlaSimulator (developed by University of Barcelona).

Monish Kanna S S

-documented by Monish Kanna for BOSCH Rexroth AG Talent Acquisition Team, Stuttgart.

You might also like