represents a substantial contribution over the past decade. The value of this approach will likely continue in the future. Translating cellular and molecular imaging modalities from the laboratory to an in vivo–in situ surgical setting will further expand the functional capabilities of surgical interventions by providing improved tissue detection, labeling, and targeting for both macroscopic and cell-based therapies. This approach can fundamentally alter the planned surgical pathways by streamlining intraoperative surgical decision making and optimization with increased consistency and accuracy, circumventing potential postoperative complications and revisions. Another way for robots to add value is through autonomy. Although the development of autonomous driving capabilities has been perhaps the hottest topic in all of robotics over the decade, the use of autonomy in medical robots is currently limited. Examples include assistive wearable robots and rehabilitation robots. These systems produce preprogrammed motions that can be switched between and altered on the basis of user inputs. Similarly, orthopedic robots mill out preprogrammed cavities in bone, and radiosurgery robots play back preprogrammed trajectories to produce the desired x-ray exposures of internal lesions. Although these preprogrammed motions represent a very simple form of autonomy, they are enabling for these applications. For example, an assistive lower leg prosthesis would be useless if the operator had to actively control the ankle motion during walking. The technological frontier in medical robot autonomy corresponds to endowing the robot with the capability to formulate and alter its plans and motions based on real-time sensor data. Examples could include autonomous laparoscopic surgery to remove cancerous lesions or autonomous transcatheter repair of a heart valve. This level of autonomy brings with it not only technical challenges but also regulatory, ethical, and legal challenges, which have yet to be fully resolved and will raise commercialization costs. Consequently, it will be much easier to incrementally add such autonomous functionality to preexisting medical robots whose value can be justified without consideration of autonomous functionality. Examples include automated suturing for laparoscopic surgery, autonomous navigation of flexible endoscopes, or autonomous electrophysiological catheter mapping inside the heart. An evolutionary trend toward progressive automation as suggested by Fig. 4 will provide time for the necessary technological developments in algorithms and sensors while allowing stakeholders time to progressively construct an appropriate regulatory and legal framework. Medical applications for which autonomy is necessary to justify the robot will be more challenging to commercialize in the short term but may be of highest value in the long term. The lower hanging fruit of this type could include simple time-critical endoluminal Fig. 4. Application-specific trend toward increasing medical robot autonomy. In current use, the level of autonomy is typically the minimum needed to be clinically useful. For example, radiotherapy robots operate at a level of conditional autonomy computing and executing a radiation exposure trajectory to provide the desired radiation dose inside a patient while minimizing exposure of surrounding tissues. Orthopedic robots are capable of autonomously milling out a prescribed cavity for knee and hip implants. In contrast, laparoscopic surgical robots have proven successful under continuous operator control and so currently offer only limited robotic assistance. Transcatheter mechanical thrombectomy and heart valve repair are examples of clinical applications for which robotic solutions have yet to be developed, although both could potentially benefit from robotic solutions. In the future, it is anticipated that the level of autonomy of current robotic systems will increase. The biggest increases will be for those applications for which autonomy is vital to their function. For example, highly autonomous systems for remotely performing emergency mechanical thrombectomies to treat stroke would substantially increase the accessibility of this treatment while also decreasing the time to treatment. As a second example, bionic implants that improve or restore body functions will be sufficiently integrated with their host to not require continuous conscious control. Downloaded from https://www.science.org on November 16, 2021 Dupont et al., Sci. Robot. 6, eabi8017 (2021) 10 November 2021 SCIENCE ROBOTICS | REVIEW 13 of 15 interventions, whereas bionic implants represent a more complicated class of devices. Of the more than 19,000 engineering papers published on medical robotics since 1990, only a handful can be considered enabling for existing commercial medical robots. Even the papers of high technological influence comprising the bibliography have modest numbers of patent citations. In part, this may be due to the substantial lag that can occur between technology development and its commercial application. Perhaps, an equally important contributor is the mismatch between technology research and the realities of medical device commercialization. Bringing robotic technology to clinical use requires much more than simply well-cited research articles. A genuine clinical need must be