This page presents the videos of the ballbot, a human-sized mobile robot that balances on a ball. The videos shown here are a result of my PhD work at The Robotics Institute of Carnegie Mellon University. The videos are arranged chronologically with the latest ones on top. Please use the side panel to navigate between them.
If you have not seen the ballbot before, please watch the introduction video first.
Graceful navigation of the ballbot in larger maps
This video demonstrates the ballbot gracefully navigating larger maps using an integrated planning and control framework. The motion planner plans in the space of controllers, and the navigation task is achieved by switching between an optimal sequence of controllers. However, the computational constraints on the robot limit the size of the map that can be used for planning in real-time. In order to navigate larger maps using this framework, the map is split into smaller regions for which the motion plans can be generated in real-time. The overall goal in the map is achieved by piecing together locally optimal sequences of controllers in the individual regions.
Surveillance and dynamic replanning
This video shows the ballbot successfully achieving a surveillance motion with ten goals using an integrated motion planning and control framework. The motion planner plans in the space of controllers, and achieves the surveillance task by switching between an optimal sequence of controllers. Although the robot is switching between controllers, the integrated motion planning and control framework ensures that the overall motion is graceful. This video also demonstrates the capability of the framework to successfully achieve a point-point navigation task in the presence of dynamic obstacles. The motion planner actively detects dynamic obstacles and replans in the space of controllers to achieve the navigation task.
Ballbot motion with additional arm constraint trajectories
This video demonstrates the ballbot's capability to achieve desired motions on the floor while subjected to additional constraint trajectories on the arm angles. The shape trajectory planner computes the acceleration that results from tracking these additional constraint trajectories and plans trajectories for the other available shape variables (body angles in this case) to both compensate for this resulting acceleration and also achieve the overall desired motion of the robot. The arm constraint trajectories shown in this video include moving the arms forward emulating a lifting operation, holding the arms forward emulating a holding operation and some asymmetric arm motions emulating hand waving or aerobics.
Graceful navigation of the ballbot
This video demonstrates an integrated motion planning and control approach that ensures graceful navigation for balancing mobile robots like the ballbot. Unlike traditional motion planning that plans in position space or path space, the motion planner here plans in control space and provides a sequence of controllers that achieve the overall navigation objective. The video demonstrates point-point motions and surveillance motions on the ballbot. The ballbot uses Hokuyo URG-04LX laser range finder and a particle filter based localization algorithm to localize itself on a map of the lab. It also uses the laser for actively detecting obstacles and avoiding them.
Ballbot motion with 2-DOF arms
This video demonstrates the ballbot's capability to use its arms for motion. A pair of 2-DOF arms were added to the robot in early 2011. The shape trajectory planner was first designed to plan motions only for the body angles as shown in the videos below. Here, the shape trajectory planner can plan in a high-dimensional shape space (space of body and arm angles) in order to achieve desired motions of the robot. The video demonstrates the ballbot successfully tracking desired motions on the floor by tracking planned trajectories for both the body angles and the arm angles. The user can choose the contribution percentage between body lean and arm motions.
Fast maneuvers with the ballbot
This video demonstrates the ballbot performing some fast, dynamic and graceful maneuvers. The motions presented here are obtained by piecing together some elementary motions called motion primitives. The piecing together operation is done such that the resulting motion is graceful with no discontinuities in position, velocity and acceleration. The motion primitives use the shape trajectory planner, which plans body angle (shape) trajectories in order to achieve certain desired motions on the floor like a straight line motion or a sharp turn.
Ballbot navigating corridors
This video shows preliminary results of the ballbot using a Hokuyo URG-04LX laser range finder to navigate corridors. The objective of this simple implementation was to actively detect obstacles and walls and keep the robot in the middle of the hallway.
Fast motions with the ballbot
This video demonstrates some of the fast, dynamic motions of the ballbot. The ballbot is an underactuated system and it must lean its body in order to move. The shape trajectory planner generates body angle trajectories, which when tracked will result in optimal tracking of position trajectories on the floor.
Human-Robot Physical Interaction
Physical interaction between humans and robots play an important role while operating in human environments. This video demonstrates some interesting human robot physical interaction behaviors with the ballbot. Thanks to its dynamic stability, the ballbot while balancing can be physically moved and stopped with very little effort, in fact a single finger is all you need to move it around. Its robustness to disturbances is again demonstrated. An interesting behavior of detecting human intentions is shown. In this behavior, a soft push to the robot is considered unintentional, whereas a hard push is interpreted as a "stay-away" command. The video also demonstrates that physical interaction can be used to teach some simple motions to the robot.
The ballbot - An Introduction
The ballbot is a dynamically stable mobile robot that balances on a single ball. It was built by Ralph Hollis, Research Professor at Carnegie Mellon University, Pittsburgh, USA. To the best of our knowledge, it the first of its kind ever built. Recently, many other groups have started building such robots. Please visit the Ballbot Wikipedia Page for more details.
My work focused on developing control and planning algorithms for the ballbot with special emphasis on enabling the robot to achieve fast, dynamic and graceful motions. This video demonstrates the ballbot's robustness and balancing capabilities.