MECHANICAL ENGINEER
Custom Robotics Joint (Neocis)
About the Project
This project focused on designing custom joints for a surgical robotic arm used in dental implant procedures. The overarching goal was to develop mechanically precise, reliable, and safe actuators capable of operating alongside human surgeons while maintaining sub-millimeter accuracy. When I joined the project, the systems team had already generated a kinematic sketch of the arm that defined the number and orientation of the joints, but no specific requirements existed at the joint level. My task was to define those requirements and design joints that would meet the needs of the robotic system.
My Role
I served as the primary mechanical designer for the joints, responsible for defining functional requirements and specifying components across the entire arm. I owned the detailed engineering drawings and full mechanical design for the two most proximal joints, which were the largest and carried the highest loads. In addition to these, I contributed design modifications and refinements to the other joints as needed, ensuring consistency and performance across the system. I collaborated closely with the electrical team on custom driver and encoder boards, and with the systems team on encoder calibration and system-level integration. I also led design reviews to iterate on mechanical concepts and validate that the joints met precision, load, and responsiveness requirements.
Functional Requirements
The joints needed to satisfy multiple requirements driven by the robotic system’s overall performance:
-
Precision: The combined robotic system (actuated guidance arm, patient tracking arm, and other error sources) needed sub-1 mm positional accuracy.
-
Load Bearing: Joints had to withstand drilling forces and forces applied by the operator.
-
Speed & Responsiveness: Must be fast enough to feel responsive, yet safe for human-robot interaction.
-
Actuator Pass-through: Hollow bore to allow daisy-chained cable routing.
To determine the load requirements for each joint, I collected a list of all clinically relevant arm poses and gathered log files from older robots currently in the field to quantify the forces applied to the drill. Using this data, I developed a Python script that randomly selected a pose and a force from the logs and calculated the resulting torque on each joint. The torque was then separated into contributions handled by the motor and by the bearings by taking the dot product along each joint axis. This process was repeated hundreds of times to determine the 99.99th percentile torque values, which were used to specify the motors and bearings.
In parallel, I developed a second Python script to analyze joint deflection. For each arm pose, the applied load was divided by the stiffness of the bearings to calculate the angular deflection, which was then converted to the positional deflection at the tip of the arm. This analysis ensured that the joint design would maintain sub-millimeter precision under all expected operating conditions.
Mechanical Design:
The final joint design integrated multiple elements to meet the precision, load, and safety requirements of the surgical robotic arm. Key components and their justifications included:
-
Frameless Brushless DC Motor: Chosen to make the actuators as compact as possible, allowing seamless integration into the robotic arm. Low-cogging motors were selected to improve smoothness and precision during operation.
-
Harmonic Drive: Selected for their zero-backlash design and ability to achieve high gear ratios. They also incorporate a high-stiffness, preloaded cross-roller bearing that handles radial and axial loads effectively. Harmonic drives provide a hollow through-bore, which allowed us to route cables through the joint center.
-
Dual Optical Encoders: One encoder was used for motor commutation and velocity measurement, while a second output encoder measured the position after the transmission. The resolution of each encoder was carefully chosen based on the joint’s contribution to the overall error budget. We aimed to select the largest possible scale for each joint to maximize precision, but this had to be balanced against the need to keep the joint compact and fit within the housing of other components. This trade-off was critical to achieving high accuracy while maintaining the small form factor required for surgical use.
-
Custom Mechanical Brake (Solenoid): A safety mechanism was required to hold the robot when power was shut down. To maintain compactness, especially in distal joints where patient visibility is important, we designed a custom solenoid brake that was smaller in the axial direction than most spring-loaded power-off brakes.
-
Wave Spring: Included to provide consistent bearing preload regardless of tolerance stack-up, ensuring predictable stiffness and minimizing variability in joint performance.
-
Machined Housing Components and Ball Bearings: These constrained all moving parts and supported the structural integrity of the joint while maintaining precision under load.
Testing & Validation
After assembly, we performed a series of tests to verify that the actuators met their torque, speed, and accuracy requirements. Loaded torque tests confirmed that each joint could deliver the expected output torque under realistic conditions, while encoder feedback validated that the actuators achieved their target speeds with stable control.
For accuracy testing, each joint was mounted to an off-the-shelf rotary table, and output encoder readings were compared against the table’s encoder as a ground-truth reference. The encoder supplier recommended spinning the output encoder at 300 RPM for eccentricity calibration, but since the harmonic drives could not be backdriven, we decoupled the encoder shaft from the output flange and spun it independently. This approach allowed calibration but introduced unwanted friction and did not perfectly represent the assembled joint’s behavior.
In hindsight, a more effective method would have been to test the fully assembled joint by mounting it to a passive high-accuracy encoder with dual readheads to directly measure and compensate for amplitude and phase errors (as was done in another later project for a different application). Despite these limitations, the joints successfully passed verification and demonstrated the expected performance across torque, speed, and positional accuracy.
Outcome
Testing confirmed that the custom joints met all key specifications, including torque output, speed, and positional accuracy, though the calibration method could have been improved. A prototype of the full robotic arm was assembled and operated as intended, demonstrating that the mechanical design was both functional and manufacturable.
Ultimately, the team decided to transition to an off-the-shelf actuator solution that was being evaluated in parallel. The commercial joints offered lower cost, faster assembly, and more consistent performance since they arrived pre-tested and calibrated. Although the custom joints were not carried into production, the project provided significant experience in actuator design, integration, and testing, and offered valuable insight into the trade-offs between custom and commercial hardware.
Lessons Learned
-
The impact of tolerance stack-up on performance: Initial designs experienced variation in bearing preload due to dimensional inconsistencies between assemblies. Adding a compliant element such as a wave spring allowed for consistent preload across units, improving performance and simplifying assembly without the need for shims.
-
The importance of considering encoder calibration early in the design process:
Optical encoders provided excellent resolution but required complex calibration and were sensitive to eccentricity. In future designs, magnetic or inductive encoders may be preferable for robustness. If optical encoders are used, incorporating dual readheads or designing a calibration setup that keeps the joint fully assembled would ensure more accurate and repeatable results. -
Concentricity and constraint design are critical for low-friction performance:
Small misalignments and over-constraints increased friction and reduced repeatability. Using higher-precision bearings with low runout (ABEC 5 vs ABEC 1) and paying close attention to concentricity during design and assembly would improve overall actuator performance and efficiency.
