11.3 - Design Process
Timeline
Although the design process involved various changes in scope, our main objective to create a jumping and walking robot dog leg remained constant from the original brainstorm. Regardless of technicalities, we knew we had to design a mechanism capable of producing the required motion profiles and force outputs while maintaining mechanical stability and repeatability.
Inspired by a Aaed Musa’s CARA robot, and Nathan Ferguson’s DINGO robot, our first design iterations involved the use of a rope-driven speed-reduction mechanism called a capstan drive. This idea was proposed with the intent of providing high torque transparency throughout the leg’s jumping stroke, and potentially simplifying the mechanism as we thought a four-bar linkage would have a more direct kinematic analysis than a five-bar or six-bar. However, the implementation of a capstan drive gave rise to non-linear, configuration dependent relationships that are difficult to model and cannot be accurately captured with standard rigid body kinematics. Because the capstan drive incorporates tensioned rope, the motion transmission of our original mechanism was governed by frictional contact and tension-dependent deformation rather than fixed geometric constraints. After multiple attempts to accurately model the relationship between our servo inputs and our desired outputs with a capstan drive, time constraints forced us to shift the scope of our design.
To replace the capstan drive, we leaned more into our DINGO robot inspiration and decided on a multi-loop planar linkage with two coupled fourbar subchains, each independently driven by their own servo. This mechanism is more well-known in comparison to our original idea, which allowed us to complete a validated kinematic analysis and fully accurate animation.
Initial Design Iterations
1. Before implementing servo dimensions, the initial design used arbitrary geometry. This iteration excluded the full capstan drive and modeled a single motor driving the hip joint. Its purpose was to visualize motion and identify the necessary linkage configuration.
2. After finalizing servo dimensions, we modeled them in Fusion and updated the mechanism geometry, using a parallelogram four-bar to link the capstan-driven input to the leg angle. During assembly, we discovered the large drum was not rotating about its center, compromising motion accuracy and drum contact. Due to downstream constraints from existing joints, this required rebuilding the assembly from scratch.
3. In the third assembly, we focused on correctly modeling the capstan input-output relationship, ensuring the large drum pivots about its center while allowing the top motor A’s shaft to pass through and drive the hip via a bearing. We also shortened the hip-to-leg link and removed the parallelogram to improve motion; however, the effective link on the large drum was too short, limiting the mechanism’s range of motion.
4. In our final design of the original mechanism, we tested the capstan by modeling coils on both drums with matching pitch, and came to the important realization that the coil pitch around the small drum and large drum should not be equal, but proportional. Additionally, the two drums should have been equal in width, which would disproportionately enlarge the volume of our capstan component relative to the leg linkage. While the capstan design contained obvious errors, this design also replaced arbitrary joint pegs with 6 mm ball bearings, improving ease of joint assembly and motion smoothness. With motion links defined between the slider–hip and the two drums, we were able to run a motion study.
5. After running into too many difficulties manufacturing the original Capstan drive based design, we pivoted to a linkage-based mechanism. We substituted the capstan drive for a second four-bar linkage (O2 A B O1), which is coupled to the existing four-bar linkage (O1 C D E) through a ternary link. Best described as a two-servo planar six-bar leg mechanism with a four-bar input stage and a coupled four-bar lower-leg stage. Combining design methods from existing versions of this mechanism, we defined each four-bar linkage with parallelogram geometry (O2 A = B O1; O1 E = CD; O1 C = D E) and optimized the individual geometry of each link for our desired motion. We imported M3x16mm and M3x20mm threaded fastener CAD files to simulate each joint, and used threaded inserts in our holes for the most accurate fabrication result. The motion study for this assembly was created by defining appropriate joint relationships, and analyzing the servo input angles to optimize the system’s output position at different points in the jumping motion. The vertical slider joint was also included in this motion study to simulate the desired height of the vertically constrained servos.
Physical Prototypes
Our first physical prototype used cardboard and toothpicks to model the geometry of our capstan design’s fourth iteration. Similar to the first design iteration, this initial prototype was designed to simply visualize the motion and determine if the assigned link lengths are feasible.
Once we abandoned our capstan design and opted for the coupled four-bar linkages approach, with 2 servo motors, one of which is controlling the hip and the other the knee joint. We originally opted for a wire-based design for better performance under jump tension but under experimentation it soon proved prone to breaking parts and was unable to lift the knee during peak jump to increase the ground clearance.
Final Prototype
For our final prototype, our team designed and built a robotic dog leg that focused on creating realistic, controlled leg motion while still being simple enough to manufacture, test, and improve. Rather than treating the leg as just an idealistic mechanism, we approached it as a complete engineering system with three major goals: accurate kinematics, repeatable physical testing, and enough structural strength to survive dynamic loading and prolonged experimentation. Our final design was not only meant to move through a walking motion, but also to help us understand how a robotic leg stores energy, transfers force, and interacts with the ground.
One of the biggest improvements in our final design was the addition of proper kinematic analysis. We modeled the leg geometry using linkage equations originally so that the motion of the foot could be predicted before testing. Instead of guessing where the leg would move, for the final version we used link lengths, joint angles, and loop-closure equations, alongside the expected servo torque output to calculate the foot position throughout the leg’s range of motion. This allowed us to compare the theoretical path of the foot with the physical motion of the prototype. By doing this, we were able to adjust the geometry of the final prints more intentionally and avoid designs that looked good in CAD but created awkward or limited motion while integrated in reality.
The finalized kinematic equations also helped us understand how changes in one joint affected the rest of the leg. Since robotic legs are highly dependent on joint coordination, small changes in link length or angle can significantly alter the foot path, especially with ground contact. Our team used this analysis to improve the leg’s range of motion and make the foot travel in a more useful path for walking and pushing off the ground. This made the final prototype feel much more like an engineered mechanism rather than a simple assembly of parts we threw together.
Another important feature of our final prototype was the jumping test stand. We built the stand so the leg could be tested vertically in a controlled environment without needing to build a full robot body. This allowed us to study how the leg behaved during compression, extension, and ground contact. The test stand gave the leg a guided path of motion, which made it less damage prone and easier to collect useful experimental results. Instead of only checking whether the leg could move, we could evaluate whether it could actually generate upward force and recover after impact.
The jumping test stand also made testing more repeatable. Since the leg was mounted in a fixed structure, we could run multiple trials and compare how the design performed under similar conditions. This helped us identify problems such as mechanical slop, binding, friction losses, and weak connection points. It also gave us a better way to test improvements because we could change one part of the design and immediately see how it affected the jump performance. This was one of the most valuable parts of the final prototype because it connected the theoretical design work to real physical behavior.
We also improved the mechanical design of the leg to make it stronger and easier to assemble. The final version used a more organized joint layout, better link hole spacing, better manufacturing, and more reliable fastener designs. We paid attention to alignment because we learned even small off planar misalignments caused extra friction and made the leg harder to move. By improving the joint supports and simplifying the assembly process, the leg became smoother and more consistent during testing. These changes made the prototype more durable and increased jump height by approximately 3 mm.
A major lesson from the final design was that robotic leg performance depends on more than just motor power. The geometry of the leg, the stiffness of the structure, the friction in the joints, and the path of the foot all affect how well the system works. The jumping test stand showed us that energy can be lost before it ever reaches the ground. Because of this, our final prototype focused heavily on making the motion controlled and repeatable, not just making a leg move.