Robotics

Skills: Robot Operating System (ROS), Sensor Fusion, Path Planning Algorithms, Control Systems, Machine Learning for Robotics, Kinematics and Dynamics, Simulation Tools (Gazebo, V-REP), Embedded Systems Programming, Computer Vision Integration, Real-Time Processing.

Carnegie Mellon University | Humanoid Robotics and Cognition

Multi-Modal Meta human

Developed a reinforcement learning model to synchronize audio and motion for a humanoid robot's dynamic behaviors in a virtual environment, achieving real-time interaction and task completion.

Reinforcement Learning for Task Execution

Designed a reinforcement model that enables the humanoid (metahuman) to navigate towards targets and perform tasks upon command. Reinforcement learning optimizes the model's responsiveness to achieve seamless goal-oriented behavior.

Multi-Modal Synchronization

Integrated audio and motion modalities within Unreal Engine 5 to align the robot’s audio cues and physical actions, creating a cohesive and realistic behavior profile. This multi-modal synchronization enhanced the robot's interaction quality in dynamic settings.

Simulation in Unreal Engine

Developed and tested the humanoid behaviors in a custom Unreal Engine 5 environment, simulating real-world conditions with physics-based controls. This setup allowed for accurate performance evaluations and adjustments for optimized motion and interaction fidelity.

Computer Vision

Skills: Image Processing Techniques, Feature Extraction Methods, Deep Learning Frameworks (TensorFlow, PyTorch), Object Detection and Recognition, 3D Vision and Reconstruction, Camera Calibration and Geometric Modeling, Optical Flow and Motion Analysis, Segmentation Techniques, Augmented Reality (AR), Robustness and Generalization.

Carnegie Mellon University | Space Robotics

IRIS: Lunar Rover Project

Assisted in developing a Lunar Rover ideation to explore autonomous localization and navigation on challenging lunar terrains, focusing on robust navigation, vision-based perception, and adaptability to lunar surface conditions.

Rover Design Overview

Engineered a modular rover with rugged wheels and lightweight materials for lunar terrain. Integrated a solar power system to support extended operational periods in simulated lunar environments.

Navigation and Localization

Implemented EKF-based state estimation with IMU and encoder data, integrating visual-inertial odometry for real-time navigation. Autonomous path planning used precomputed lunar maps to optimize paths.

Camera and Perception

Used stereo vision cameras for terrain mapping and obstacle detection. Applied depth estimation and image processing for 3D point cloud generation, enabling precise navigation on low-contrast lunar surfaces.
     Back to All Projects
Sushmith Thuluva
Connect with me