Basanti is an interactive animatronic robot that is inspired by the human face itself. Equipped with the basic attributes of the human face, like speech, vision, motion, expressions and other facial parts; it is the complete framework that merely needs a synthetic face mask to be a human face lookalike.
Developed at Raman Lab - Department of Electrical Engineering, Malaviya National Institute of Technology Jaipur, under the mentorship of Dr. Rajesh Kumar, Basanti 1.0 is the first of its kind in-house developed actroid with the project made from scratch at our own facility.
REPLICATING THE FACIAL MOVEMENTS
Creating a complete replica of human face meant it has all the motions of the human face
The robot has multiple motions, each supported by an underlying custom designed mechanism powered by closed-loop servo motors. Aimed at precision in movement and accuracy in the same, without generating much a backlash; all the joints are provided little higher than rated power to ensure maximum efficiency of the motors. Modular structure reduces the dependence any one mechanism over the other.
The movable joints of Basanti 1.0 include:
Upper & Lower Lips
One servo motor drives a two-spur gear arrangement(1:1 gear ratio) with one half of the lip mounted atop each spur gear and make each half converge and diverge simultaneously, just like human lips. The upper lip arrangement is attached to the immovable face by suspending the motor assembly from the rear while the lower lip arrangement is housed inside the jaw which is itself movable - just like a human face.
Eyeballs
3D printed rack and pinion arrangement driven by one servo motor has the two eyeballs move two and fro by attaching extensions to the rack
Jaw
Another powerful servo motor powers this jaw revolute joint that also has the lower lip motion arrangement. It is the most important intricate part of Basanti 1.0 that gives it a human-like appearance.
Neck - Pan and Tilt
Pan motion is facilitated by a high torque dual-shaft servo motor that is attached to the complete setup. Tilt motion is difficult to achieve and needs extremely high torque. To power this revolute joint, a geared-DC motor with custom-designed feedback setup is used in order to reduce costs and the required input power.
Basanti Physical Movements
FACE DETECTION AND RECOGNITION
Visual feed isn’t just limited to seeing the environment, but also extracting the details out of it
Basanti 1.0 does perform object detection and identification, but the primary task means detecting and recognising human faces out of it. Using python implementation of face recognizer descibed in paper “FaceNet: A Unified Embedding for Face Recognition and Clustering” ,Basanti 1.0 is capable of facial identification with as less as one training image. Further versions should test in-house developed techniques for better real-time performance that doesn’t have the limitations of frontal view/side view or picture quality.
Face Detection & Recognition
FACE TRACKING
A human face is capable of understanding the environment visually and perform related tasks
Basanti 1.0 has a webcam mounted on its head that acts as its eyes, giving it the visual feed of the surroundings. However, usage of single camera doesn’t allow for stereovision which is the primary requirement for depth-perception and focussing- one of the major tasks of the human eye.
Face tracking task achieved by Basanti 1.0 currently shows only the pan motion of the neck to the track the faces in front and keep the face aligned to the centre of its canvas. Future versions of Basanti are envisaged to have multiple axis tracking as well capability of doing the same with the eyes.
Face Tracking
BASIC QUESTION-ANSWERING
Listening, understanding and then replying to one’s questions forms the foundation of communication in human beings
Basanti 1.0 has a mic as well as a speaker that act as the ears and the voice of the robotic face. Running a CNN on the inside, Basanti 1.0 is capable of listening to the user’s questions, processing them via various algorithms - from basic keyword search to advanced end to end memory networks and then replying with an answer derived from either the biograpy fed to it or it deduced from the internet via the operational algorithms.
Speech Based QA
And finally, the team behind all this
