The next major study using the robotic transoral lung system presented in Chapter 4 is already underway, and it involves the integration of the system with an existing image- guidance system and testing with more anatomically realistic tissue [79]. Preliminary re-
sults are promising, with the system able to achieve less than 2 mm targeting error in inflated, ex vivo porcine lung using a motion planner for obstacle avoidance [79]. Further experiments will be necessary, including the use of perfused tissue and respiratory mo- tion. The use of perfused porcine tissue will allow the flexure-tip steerable needle to be optimized for lung parenchyma. This is likely to result in a larger achievable workspace for the needle, thus enabling more complex motion planning problems to be investigated [140, 141]. Methods to compensate for respiratory motion using the robotic lung system should also be developed, as the system is ultimately intended for use on patients who are breathing on their own. Breathing compensation could be addressed with a robotic actua- tion mechanism that monitors the patient’s chest and moves the entire system such that it is in sync with the chest wall. Another approach is to only advance the system when the patient is in a breath hold, essentially removing the effect of respiratory motion.
Another open area of research for the lung system involves optimizing the design of concentric tube robots based on anatomical constraints. There is a large design space avail- able to concentric tube robots, and work is being done to answer the challenging question of optimal tube design [141, 187, 214, 215]. A critical avenue for future work with re- spect to the robotic lung system is the development of a new actuation unit. The current actuation unit was not designed with compactness in mind. The next version of the robotic actuation unit should be reduced in size such that it can be easily managed by the physician in the operating room. Although it is envisioned that the physician will deploy the bron- choscope manually in the system workflow, it may be useful to investigate roboticizing bronchoscope deployment in the future (see e.g. [43]), particularly if a robotic system can maneuver the scope more accurately and reliably than a human. However, it will likely be desirable from a clinical adoption and cost standpoint to rely solely on manual deployment which will minimally impact the established workflow of bronchoscopy. It is also worth noting that our system could be used in conjunction with virtual bronchoscopy [124] to assist the physician with manually navigating the bronchial tree, and this integration would
be a worthwhile avenue of future research.
The robotic endonasal system developed in Chapter 5 also presents a number of areas for continued investigation, and some of these avenues are currently underway. First, a new actuation unit has been designed and is currently being assembled that considers the requirements of an operating room such as sterility and compactness. While we have de- signed robots for sterility and safety in the past [188], the new system must also consider interchangeability of the tools, positioning of the robot over the patient, and redundant sens- ing in the event of an emergency. These aspects will need to be integrated with the next version of the actuation unit. Another avenue for future work that would benefit not only our system but also any systems or tools developed for the skull base is the investigation of forces encountered during skull base procedures. In [148], we performed a preliminary study to look at the forces experienced by a surgical curette during endonasal pituitary tumor removal. An extension of this study could collect similar data for additional pitu- itary removal cases or investigate tool tip forces during various other endonasal procedures.
Additionally, investigating the forces required to break different bony structures using ca- daveric specimens would be useful. Such force information provides maximum limits for force application which can then be integrated into a robotic system such as ours or could be used for training the next generation of surgeons.
If this testing reveals that some areas of the skull base are sensitive to anatomical varia- tion such that a safe operating threshold cannot be developed, the implementation of colli- sion avoidance algorithms [211] or “no-fly zones” [216] can be integrated with the robotic system. There are also many open questions with regards to the optimal user interface for continuum robots. While touchless user interfaces have been shown to be one possibility for use with the robotic endonasal system [217], many other approaches and interfaces can be imagined, and a thorough investigation into the optimal user interface for continuum robots would be valuable. Experimentally, the next steps for the endonasal system are to conduct more rigorous cadaver studies using the newly designed quadramanual actuation
unit with suction provided through one of the arms, allowing for complete removal of pi- tuitary tumors with the system. While performing cadaver studies, it will also be useful to investigate robust calibration methods for use during a procedure, in addition to calibration of the system before beginning the operation. Toward this end, it would be good to investi- gate calibration using collisions on physical hardware using the new quadramanual system.
In order to generate collisions on physical hardware, the same resolved-rates method used in Sec. 5.6.3 can be used, or the collisions can be induced by a user teleoperating the sys- tem. One approach to performing calibration using collisions is to construct a circuit that measures the electrical resistivity of the tubes in order to record the total arc length along each robot backbone to a collision point. As shown in Sec. 5.6.3, calibration is able to be performed using the total arc length along each robot backbone to the collision point.
Another calibration approach using collisions is to determine the exact moment when the robot arms collide and use statistical estimation to infer the current model shape of the robots and the most likely location of the collision point and arc lengths [212]. Rather than relying on electrical resistivity to measure the arc lengths (which could be noisy), this approach uses the certainty provided by a physical collision coupled with statistical estima- tion to obtain the arc lengths along the robot backbones to the collision point. In the course of these experiments, calibration using collisions can be investigated in the presence of ac- tuator uncertainty and uncertainty in the initial configuration of the robot. Finally, a more robust calibration can be generated by incorporating a priori parameter estimates [204].