Research Projects

The efficacy of performing yoga asanas is highly correlated with the accuracy of limb movement. A recent study by the University of Sydney found that around 10% of people will develop injury or pain from yoga, comparable to the injury rate from all other sports activities in a similar population. This implies that there is still a cent percentile possibility to do yoga asanas incorrectly. To ensure the efficacy of performing yoga asanas, an intelligent system for limb movement prediction is essential. Three novel techniques are used to predict limb movement: using commercially available analog-output Inertial Measurement Unit (IMU) sensors and an indigeneous mixed signal neural network chip. The proposed ASIC contains an analog neural network with programmable weights and biases that can be reconfigured based on digital values. The neural network may be a multi-layer perceptron that detects certain signal artefacts in the IMU output. If particular signal artefacts match with known ones, a digital representation of the artefacts will be stored with a timestamp in an on-chip buffer. A low-cost energy-efficient small microcontroller IC can be placed on a PCB with the proposed ASIC. This microcontroller can read the values of artefacts resulting for a particular limb movement from the ASIC buffer and transmit these values wirelessly to a central control and analysis AI platform. This approach reduces energy consumption, increases wireless components' reliability, and reduces algorithmic complexity of the microcontroller firmware on the band. The main objectives of the proposal are to design and develop CareNet, a neural network architecture based on IMUs for 3-Dimensional activity recognition in yoga, use the Google Corel Dev Board to prototype the embedded system, and fabricate Analog Deep Neural Network ASIC for compact, low power high precision limb movement tracking and prediction. The methodology in phases of six months duration includes literature survey, initial data collection for yoga asanas, developing circuit modules, preparing analog signal datasets, developing preliminary software models, developing a Recurrent Neural Network mixed-signal backend, studying vibration feedback, completing the DRC and ASIC layout, starting the development of AI-based full body motion tracking and feedback system software, fabricating PCBs for ASIC testing, and building and testing wearable and CCAU devices with the help of a yoga master, trainer, and trainees.