I'm a fourth-year Robotics Ph.D. student at Georgia Tech, advised by Dr. Sehoon Ha and Dr. C. Karen Liu. I am interested in applications of vision-based robot learning in real-world robotics. Currently, I am working on outdoor navigation and environment interaction problems.
We present a learning-oriented morphology optimization framework that accounts for the interplay between the robot's morphology, onboard perception abilities, and their interaction in different tasks. We find that morphologies optimized holistically improve the robot performance by 15-20% on various manipulation tasks, and require 25x less data to match human-expert made morphology performance.
[project page] [video overview] [pdf] [arXiv]We propose a novel motion control system that allows a human user to operate various motor tasks seamlessly on a quadrupedal robot. Using our system, a user can execute a variety of motor tasks, including standing, sitting, tilting, manipulating, walking, and turning, on simulated and real quadrupeds.
[project page] [pdf] [arXiv] [video]We introduce Bootstrap Across Multiple Scales (BAMS), a multi-scale self-supervised representation learning model for behavior analysis. We combine a pooling module that aggregates features extracted over encoders with different temporal receptive fields, and design latent objectives to bootstrap the representations in each respective space to encourage disentanglement across different timescales.
[project page] [pdf] [arXiv]We design a system which enables zero-shot vision-based policy transfer to the real-world outdoor environments for sidewalk navigation task. Our approach is evaluated on a quadrupedal robot navigating sidewalks in the real world walking 3.2 kilometers with a limited number of human interventions.
[project page] [pdf] [arXiv] [video] [TechXplore article]We train vision-based agent to perform object searching in photorealistic 3D scene. And propose a motion synthesis mechanism for head motion re-targeting. Using which we enable object searching behaviour with animated human character (PFNN/NSM).
[pdf] [arXiv] [video] [talk(20 min)]We show how vision-based navigation agents can be trained to adapt to new sensor configurations with only three shots of experience. Rapid adaptation is achieved by introducing a bottleneck between perception and control networks, and through the perception component's meta-adaptation.
[pdf] [arXiv]Image domain adaptation through the conversion of images with randomized textures (or real images) to a canonical image representation. Replication of a RCAN paper with different loss modeling (Perceptual/Feature Loss instead of GAN loss).
[github]Computer Animation class project, which utilizes off-the-shelf Soft-Actor-Critic Reinforcement Learning method that learns to build up the momentum and swing the animated character on a pull up bar.
[short-summary]Mobile Manipulation project that utilises MoveIt! & GQ-CNN to grasp an object from the table using a Fetch Robot in the Gazebo Simulator.
[short-summary]End-to-end (image-to-steering wheel) control policy learning from data collected over multiple laps with off-the-track recoveries generated by human.
[github]
2023© Maks Sorokin
built using Skeleton,
icon credits flaticon,
hosted by GitHub Pages❤️
feel free to copy: this page