Shankara Narayanan Vaidyanathan

In this work, we've developed a non-linear optimization based system that tightly integrates GPS data, inertial measurments and visual input from a multi camera setup. Building upon our Generic Visual SLAM framework for multi-camera setups, we utilize IMU preintegration to summarize hundreds of inertial measurements into a single relative motion constraint. For achieving precise global localization, we've developed a custom GPS factor. This factor allows for real-time estimation of the transformation between the Visual-Inertial Odometry (VIO) frame and the GPS's ENU frame while accounting for the time offset between the sensors.

In this work, we present a generic sparse visual SLAM framework capable of running on any number of cameras and in any arrangement. Our SLAM system uses the generalized camera model, which allows us to represent an arbitrary multi- camera system as a single imaging device. Our system can adapt to different camera configurations and allows real-time execution for typical robotic applications.

We worked on developing an end-to-end framework to solve the task of table-top rearrangement planning in multiple scenarios, from as simple as placing the objects into a box, to arranging huge books onto a book holder. This involved working on pose estimation, grasp pose detection, task sequence planning, and motion planning to demonstrate the framework’s applicability to several challenging rearrangement problems on a variety of objects belonging to the YCB dataset and other unknown objects that represented real-life objects.

In this work, we approached the problem of navigation of Micro Air Vehicles (MAVs) amongst urban high-rises with a single monocular camera as the essential sensing modality. We propose a novel flow synthesis based visual servoing framework enabling long-range obstacle avoidance for Micro Air Vehicles (MAV) flying amongst tall skyscrapers. We showed tangible performance gain in comparison with popular flow-based obstacle avoidance methods, and depth-based obstacle avoidance pipelines on both photorealistic and real-world scenes imported into the AirSim Simulation Environment

Generally, we can divide the approximations used in path planning into 2 types: Search-based and Sampling-based.
A recent approach called Batch Informed Trees (BIT*) combines the strengths of both search-based and sampling-based planners. In this work, we have used the pseudo-code from the paper and coded the algorithm from scratch, and tested its performance in R2 space for different motion planning scenarios using a custom visualizer.

State Estimation: Issues During Indoor-Outdoor Transitions

Understanding the state of the robot is a crucial element for autonomous navigation. Several papers investigate SLAM (Simultaneous Localization and Mapping) pipelines in either outdoor or interior contexts, or both. However, only a few works have looked at the difficulties that might develop when moving from an outdoor to an indoor setting or vice versa. In this project, we have explored some frequent problems faced when performing global state estimation during such environment transitions. This project used three key sensors: Stereo Camera, IMU, and RTK-GPS.

This assignment involved building a basic navigation stack using two sensors: a BU-353 GPS and a VectorNav VN-100 IMU. The main focus was on calibrating and analyzing the IMU data. The GPS and IMU data were collected by using the NUANCE car and driving it around the campus. Magnetometer Calibration was done and then this data was fused with gyroscope data using a complementary filter, providing a more accurate estimate of the sensor's orientation. A dynamic bias correction for the accelerometer was done using GPS velocity. The filtered yaw and position estimates were used to perform dead reckoning.

Occupancy grid maps provide a convenient representation of a robot’s environment. This assignment involved coding up A* and Probabilistic roadmap (PRM) path [planning algorithms for a binary 2D occupancy grid generated using the Cartographer SLAM system.