2025 Fall VIP Projects – Laboratory for Intelligent Decision and Autonomous Robots

Quadrupedal Navigation [quadnav]

Mentor: Max Asselmeier (mass@gatech.edu)

Project description: The quadrupedal navigation (QuadNav) project will continue our ongoing efforts to establish a pipeline for online, real-time perception-informed navigation using the Robot Operating System (ROS). This navigation pipeline will include global and local planning, a simultaneous localization and mapping (SLAM) process, and other quadruped-specific modules including terrain traversability and steppability estimation.

Ideal team size: Four, roles would include (but are not limited to): SLAM, Perception, Planning, Control.

Expected Outcomes:

For this upcoming semester, we will establish a systematic testing set-up for hardware navigation trials and perform bi-weekly navigation experiments where we can tune and evaluate our pipeline.

At the end of the academic year, we will deploy steppability and traversability models onboard our NVIDIA Jetson Orin NX PC on our Unitree Go2 Quadruped to guide our navigation planner.

Necessary skills: Linux, C++, Python

Preferred skills: Robot Operating System (ROS)

Grading Info: Grading will be based on attendance, communication, and participation.

Reference:

Planners that we aim to deploy the quadrupeds on are similar to the following:

[1] Potential Gap: A Gap-Informed Reactive Policy for Safe Hierarchical Navigation. Ruoyang Xu; Shiyu Feng; Patricio A. Vela. 2021. Link: https://ieeexplore.ieee.org/document/9513583.

[2] GPF-BG: A Hierarchical Vision-Based Planning Framework for Safe Quadrupedal Navigation. Shiyu Feng; Ziyi Zhou; Justin S. Smith; Max Asselmeier; Ye Zhao; Patricio A. Vela. 2023. Link: https://ieeexplore.ieee.org/document/10160804.

[3] Dynamic Gap: Formal Guarantees for Safe Gap-based Navigation in Dynamic Environments. Max Asselmeier, Ye Zhao, Patricio A. Vela. 2022. Link: https://arxiv.org/abs/2210.05022.

Ultimately, we want to develop an autonomy framework that could be deployed for tasks such as the DARPA SubT Challenge and the ICRA Quadruped Robot Challenge:

Steppability:
[4] Steppability-informed Quadrupedal Contact Planning through Deep Visual Search Heuristics. Max Asselmeier, Ye Zhao, Patricio A. Vela. 2024. Link: https://arxiv.org/abs/2501.00112.

Traversability:
[5] Wild Visual Navigation: Fast Traversability Learning via Pre-Trained Models and Online Self-Supervision. Matías Mattamala, Jonas Frey, Piotr Libera, Nived Chebrolu, Georg Martius, Cesar Cadena, Marco Hutter, Maurice Fallon. 2024. Link: https://arxiv.org/abs/2404.07110.

Humanoid Loco-manipulation Skill Learning [locoman]

Mentor name: Zhaoyuan Gu (zgu78@gatech.edu)

Project Description: In this project, we will explore the state-of-the-art machine-learning approaches that enable our humanoid robot to perform a series of locomotion and manipulation tasks, such as pushing through a spring-loaded door. Instead of traditional reinforcement learning, we will explore the diffusion model, an imitation learning technique that has been shown to achieve versatile skills.

Digit Humanoid Robot Loco-Manipulation and Its Integration with a Third Arm [arm]

Mentor name: Fukang Liu (fukangliu@gatech.edu)

Project Description: This project explores the capabilities of humanoid robots enhanced with supernumerary limbs. Potential tasks include opening doors while carrying a box, interacting with overhead areas, and using the extra arm as a contact point for extreme movements. The project will leverage diverse datasets (e.g., from model-based methods, model-free methods, motion capture, and video) to develop and refine the locomotion and manipulation skills of the Digit humanoid robot and its third arm. Additionally, the project aims to attach grippers to all three arms and investigate sequential collaborative tri-arm loco-manipulation tasks. Simulation will be conducted in IsaacLab, followed by testing on hardware. The scope includes both software development (policy training and deployment) and hardware design (third-arm prototyping and control).

Reinforcement Learning for Humanoid Robots [rl]

Mentor name: Feiyang Wu (feiyangwu@gatech.edu)

Project Description: This project will train and deploy Reinforcement Learning agents on humanoid robots. We will develop our own training algorithm and use IsaacLab to train RL agents. We will conduct extensive hardware experiments on our Digit robot for locomotion tasks.

Bipedal Navigation over Challenging Terrain [hector]

Mentor name: Ziwon Yoon (zyoon6@gatech.edu)

Project Description: Recent advances in bipedal robotics have driven significant progress, yet traversability and autonomous navigation for bipedal locomotion remain underexplored. Although bipedal robots have distinct advantages, including the ability to navigate diverse terrains and manipulate objects, they are more prone to instability and falls compared to mobile or quadrupedal robots. To address these challenges, we developed a Stability-Aware Traversability Estimation and Navigation framework for bipedal robots. Its learning-based traversability estimator helps the navigation stack to plan optimal paths that keep the robot within the desired stability limits. We validated it through data collected on challenging terrains—such as rough, sloped, and deformable surfaces—demonstrating its effectiveness in both simulated and real-world environments.

Whole-Body Control for Human-Humanoid Collaborative Transport Task

Mentor name: Jaehwi Jang (jjang318@gatech.edu)

Project Description:

The goal of this research is the development of a whole-body control method for humanoid robots for collaborative transportation tasks with humans. Most recent developments in real-world humanoid control have focused on single-agent motion in open environments. Our work extends these capabilities to support dynamic, responsive teamwork in complex, interactive environments.

Our approach combines imitation learning with social skill learning to enable robots to effectively coordinate with humans. We are developing the Sim2Real framework for transfer to a real-world collaborative task.