I experimented on methods to reduce the high computational power requirement of deep-learning based light field saliency detection algorithms and developed fast and accurate light field saliency detection algorithm with low computational complexity, leveraging on RGB saliency detectors. Our model demonstrated state-of-the-art performance compared to light field saliency detection models in terms of speed with comparable and better Fβ values
We developed two robotics platforms collaborating with each other to carry out the necessory tasks in the arena. You can check the video of the first round of the competition here 
We developed traffic sign, traffic light, lane and road marking detection algorithms suitable for chaotic and unstructured road scenarios in Sri Lanka and created the first large datasets for traffic sign, traffic light, and road marking detection for Sri Lanka, containing unique challenges in a developing country with scenarios covering traffic, rain, dazzle light, and normal conditions. Our row-wise classification-based lane marking detection algorithm which outperforms state-of-the-art in terms of speed with comparable and better F1 values and the end-system was optimized for real-time performance in Nvidia-Jetson Xavier with ROS.You can check the video of the final system here here.
We developed a robotic platform with line and wall following, grid solving, and coin collecting capabilities with a 1 DoF arm to carry out the tasks in the arena. This was carried out as a part of the undergraduate module: Robot Design and Competition. You can check the video of the final demo run here. 
We developed a robotic platform with line and wall following, grid solving, object detection, and colour detection capabilities with a mechanical gripper to carry out the tasks in the arena. You can check the video of a trial run here. 
Designed an application specific processor with a custom ISA for processing RGB images using DE2-115 development board. Final processor could downsample an image by an integer factor up to 15 using Gaussian and average filtering and it could apply any linear separable filter to images.UART transceiver, developed using Verilog was used as the communication medium between the processor and the PC. You can check the report and the code here.
We developed a function generator device using analog circuits to generate given waveforms (sine, sawtooth, square, and triangular). Final device had the capability to produce the waveforms in the range 13-22800 Hz, with 0-20V amplitude and it was able to provide a current up to 0.2 A.. You can check the report and the datasheet here and schematics here.