Research

Lely Center, Maassluis

Generation of photorealistic scenes using vid2vid & pix2pixhd approaches.

Generating photorealistic simulated scenes from the video/images for training autonomous agricultural robots at Lely using vid2vid and pix2pixhd approaches. We use a generative adversarial learning framework while synthesizing the simulated scenes from semantic and instance maps. These simulated scenes could be further used forimproving the training and testing of robots. [on-going project]

Robert Bosch Center for Cyber Physical Systems, IISc.

Learning manipulation tasks using Inverse Reinforcement learning methods

Current research relates to the fusion of machine learning and control theory in the domain of manipulation. Inverse reinforcement learning techniques(like generative adversarial imitation learning) & deep reinforcement learning methods have a huge advantage of leveraging prior human demonstrations to acquire seemingly simple control tasks (like peg-in-hole insertion or even a simple pick and place) for realizing a wide variety of complex behaviors in nonlinear systems.

Computational Intelligence Laboratory, Aerospace Engineering Division, IISc.

DDPG

Bipedal walking robot using Deep Deterministic Policy Gradient

The autonomous stable walking of the planar bipedal walker in simulation environment is achieved using Deep Deterministic Policy Gradient(DDPG) based on an actor-critic learning framework [for continuos action space]. The robot demonstrates successful walking behaviour by learning through several of its trial and errors, without any prior knowledge of itself or the world dynamics. Read more

Advanced Flight Simulation Laboratory, Aerospace Engineering Division, IISc.

UAV

Q-Learning for autonomous navigation of UAVs in indoor environments

Indoor Path Planning and Navigation of an ArDrone based on simple PID control+Q-Learning. The quadrotor learns to manoeuvres towards the goal point, along the uniform grid distribution(5X5) in the simulation environment based on a reward of -1 while not ending up in the goal state and a reward of +100 for reaching the goal. Read more

Projects

ABB Robotics Motion Division, Bangalore

Anthropomorphic Robotic Arm(ARA)

A.R.A. - acronym for Anthropomorphic Robotic Arm aims at implementing a collaborative robot environment for various industrial automation process. A 3D printed humanoid arm(with 5 DOF) is attached to the ABB IRB 1600 ID industrial robot to improve its dexterity and manupulation abilities. Actuation of robotic fingers is achieved using servomotors controlled using an Arduino Uno microcontroller which is interfaced to the ABB IRC5 controller via a custom designed step-down voltage regulator circuit board. Read more