Designed by Anosh Sethna. Proudly created with


During my graduate studies at UC Berkeley I worked within the Berkeley Emergent Space Tensegrities (BEST) lab to further research surrounding the application of 6-bar tensegrity robots as rovers for extraterrestrial exploration. This research was enabled through funding from NASA's Early Stage Innovations (ESI). I worked within a two-person group on the design, manufacturing, and testing of a gimbaled thruster testbed. This testbed (shown on the left) was designed to gather meaningful data regarding the stability and viability of proposed thruster-based robot locomotion.  


I acted as the Project Lead and Design Lead for the motorized gimbaled thruster, structural frame, and wire harness routing portions of our testbed. Additionally, my teammate and I worked collaboratively to integrate the sensing components of the testbed and employ a data acquisition system to collect sensor data during thruster tests.


Identify Key Requirements


Concept Selection


Testing and Analysis


6-bar tensegrity robots (see picture on the left) utilize a network of rigid structural rods and tensioned cables to create a compliant spherical geometric structure. The robot is able to shift its center of gravity by actuating various cables, which subsequently induces a rolling motion. This rolling, combined with a centrally-mounted thruster for hopping, characterizes the locomotion strategy employed by the robot. 

Our testbed was developed with the intention of creating a platform that can be used to understand the system dynamics associated with a centrally mounted gimbaled thruster and provide insights into the viability of thruster-based hopping. We started our design process with collaborative discussions to identify the key functional requirements of the testbed. The results of these discussions are summarized in the table below. Overall, the requirements centered on creating a testbed that allowed us to mimic thrust events that were previously modeled via simulations at the BEST lab, and collect data characterizing the dynamic response of the gimbaled thruster and cable-based gimbal attachment scheme. With our functionality requirements in mind, we moved on to the ideation stage of our design process.

As mentioned above, I acted as the design lead for the gimbaled thruster portion of our testbed. When developing the gimbaled thruster I went through a series of ideation sessions to aid in defining each subsystem of the overall architecture. I split the gimbaled thruster into three key subsystems - thruster, gimbal architecture, and gimbal motors. The picture on the right below helps illustrate one brainstorming session, which focused on building out a list of potential gimbal architectures. Each brainstorming session was coupled with a concept selection phase through which the top potential design solutions were scored with respect to relevant performance criteria. As an example, results of the gimbal motor concept selection are summarized below.

As  shown in the table above, servo motors provided the most optimal motor solution due to their ability to provide a relatively compact form factor with high resolution and simple position control. DC brushless motors were also considered as a compelling solution, but the added complexity associated with integrating an encoder for position feedback resulted in the decision to move forward with servo motors. Similar ideation and concept selection efforts were employed to fully define the remainder of the gimbaled thruster and the overall thruster testbed. The figures below illustrate the final testbed architecture that resulted from these design efforts. 

The structural steel frame mimics the geometry of a 6-bar tensegrity robot (see blue bars) while providing a rigid structure that isolates the dynamics of the gimbaled thruster and its attachment cables.

The gimbaled thruster uses servo motors to provide pan and tilt angle control. A disposable solid fuel thruster allows for thrust inputs ranging from 30N to 100N. A centrally mounted IMU provides gimbal acceleration data.  

Slide pots mounted in series with spring-cable gimbal fixturing mechanisms allow for displacement/tension data collection. A pulley system minimizes energy losses due to friction.

The hardware and sensor assemblies outlined above were fabricated and assembled by hand using resources offered by the student machine shop. The completed system was then integrated with a National Instruments MyRIO controller and LabVIEW GUI to enable cable tension data acquisition and to control the gimbal servo motors during thruster tests. An auxiliary Arduino UNO was also used to interface with the IMU and collect acceleration data during testing. With our hardware, motor control, and data acquisition systems in place, we were able to perform field testing with our testbed. The results of these tests and preliminary data analyses are discussed below. 


The results of our fully assembled thruster testbed and field testing are illustrated in the pictures and videos below. We completed three distinct thrust tests using 30N thrusters - vertical thrusting, constant 20-degree angle thrusting, and dynamic sweep thrusting. This allowed us to analyze how the gimbaled thruster and its six fixturing cables responded to a variety of thrust trajectories. Videos of each thrust test were captured using a high speed camera and are shown below at 0.5x speed.

Vertical Thrust Test

Constant 20-Degree Angle Thrust Test

Dynamic Sweep Thrust Test

Data collected during each thrust test was plotted and analyzed using Matlab. The plots below illustrate the data that was gathered during the vertical thrust test. As shown in the first plot, the cable tension data clearly mirrors the gimbal movement observed in the high-speed video footage - an initial thrust peak followed by a segment of constant but lower magnitude thrust, and finally ending with a return to the initial gimbal equilibrium position.

The center plot resolves the cable tensions to the resulting X, Y, and Z forces. As shown by the plot, the six cable attachment scheme is successful in evenly distributing input thrust and avoiding significant net forces in the X and Y directions. The final plot illustrates the acceleration data gathered from the IMU. It is clear from the plot's sharp peaks that the 100 Hz data is not granular enough to fully capture the nominal accelerations felt by the gimbal. The data does however, provide qualitative insights into the timing of acceleration impulses

felt by the gimbal, which match the tension transitions illustrated in the first chart.  Calculating the max recorded net force shows that our testbed exhibited roughly 10% error when compared to the theoretical max thrust of the solid fuel thruster. This moderate system error is representative of energy losses due to friction, cable misalignment, and any variation in thruster performance.

The data analysis discussed above was also completed for the constant 20-degree angle and dynamic sweep thrust tests. Overall, the results demonstrated that the six cable attachment scheme was successful in producing stable system responses to varied thrust inputs. The dynamic sweep thrust test however, showed that noticeable oscillatory gimbal responses can occur with high frequency thrust angle oscillations. These results provide useful insights into the dynamic stability of a centrally-mounted gimbaled thruster and offer quantitative data that can help refine simulation parameters used by researchers at the BEST lab.

Further testing is required to fully understand the dynamics associated with the gimbaled thruster. Other independent variables including gimbal attachment architecture and attachment spring stiffness must be varied and tested to validate simulated tensegrity architectures and explore the dynamic limits of the system. Future testing will focus on exploring these variables by testing a 4-cable gimbal attachment scheme and increasing thrust input magnitudes.