Tomson Qu

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2025-150

August 11, 2025

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-150.pdf

Hexapod robots are useful for carrying out tasks in cluttered environments since they are stable, compact, and lightweight. They also have multi-joint legs and variable-height bodies, making them suitable candidates for tasks such as stair climbing and squeezing under objects in a typical home environment or an attic. Expanding on previous work on joist climbing in attics, we train a legged hexapod equipped with a depth camera and visual-inertial odometry (VIO) to perform four key locomotion tasks: (1) climbing stairs, (2) avoiding obstacles, (3) squeezing under obstacles such as a table, and (4) squeezing between narrow vertical gaps such as between furniture.

Our policies are trained with simulation data only and are deployed on low-cost hardware that does not require real-time joint state feedback. We train a teacher-student model in two phases: In phase 1, we use reinforcement learning with access to privileged information such as height maps and joint feedback. In phase 2, we use supervised learning to distill the model into one with access to only onboard observations, consisting of egocentric depth images and robot pose captured by a tracking VIO camera. By manipulating available privileged information, constructing simulation terrains, and refining reward functions during phase 1 training, we are able to train the robot with skills that are robust in non-ideal physical environments. We demonstrate successful sim-to-real transfer and achieve high success rates across all four tasks in physical experiments.

We have integrated an adjustable-height camera mount that enables the robot to squeeze under low obstacles such as a table or a sofa. This additional degree of freedom allows our policies to automatically reduce the robot’s height by lowering the camera making it more compact during squeezing tasks. This enables the robot to squeeze under realistic furniture at home, such as a table or sofa.

Advisors: Avideh Zakhor

---

BibTeX citation:

@mastersthesis{Qu:EECS-2025-150,
    Author= {Qu, Tomson},
    Title= {Locomotion Skills for Reconfigurable Hexapod Robots},
    School= {EECS Department, University of California, Berkeley},
    Year= {2025},
    Month= {Aug},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-150.html},
    Number= {UCB/EECS-2025-150},
    Abstract= {Hexapod robots are useful for carrying out tasks in cluttered environments since they are stable, compact, and lightweight. They also have multi-joint legs and variable-height bodies, making them suitable candidates for tasks such as stair climbing and squeezing under objects in a typical home environment or an attic. Expanding on previous work on joist climbing in attics, we train a legged hexapod equipped with a depth camera and visual-inertial odometry (VIO) to perform four key locomotion tasks: (1) climbing stairs, (2) avoiding obstacles, (3) squeezing under obstacles such as a table, and (4) squeezing between narrow vertical gaps such as between furniture.

Our policies are trained with simulation data only and are deployed on low-cost hardware that does not require real-time joint state feedback. We train a teacher-student model in two phases: In phase 1, we use reinforcement learning with access to privileged information such as height maps and joint feedback. In phase 2, we use supervised learning to distill the model into one with access to only onboard observations, consisting of egocentric depth images and robot pose captured by a tracking VIO camera. By manipulating available privileged information, constructing simulation terrains, and refining reward functions during phase 1 training, we are able to train the robot with skills that are robust in non-ideal physical environments. We demonstrate successful sim-to-real transfer and achieve high success rates across all four tasks in physical experiments.

We have integrated an adjustable-height camera mount that enables the robot to squeeze under low obstacles such as a table or a sofa. This additional degree of freedom allows our policies to automatically reduce the robot’s height by lowering the camera making it more compact during squeezing tasks. This enables the robot to squeeze under realistic furniture at home, such as a table or sofa.},
}

EndNote citation:

%0 Thesis
%A Qu, Tomson 
%T Locomotion Skills for Reconfigurable Hexapod Robots
%I EECS Department, University of California, Berkeley
%D 2025
%8 August 11
%@ UCB/EECS-2025-150
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-150.html
%F Qu:EECS-2025-150