CAREER: Safe Autonomy for Soft Robots

由柔软的橡胶材料制成的机器人可以在刚性机器人无法做到的情况下与人类密切合作，因为无意的接触会减少伤害。然而，希望软体机器人在人类周围更安全的想法忽视了机器人的任务可能是多么的多面性，以及安全如何不仅仅取决于减少意外力。例如，辅助医疗程序的软机械臂必须满足对患者身体压力过大和压力不足的限制，以及其动作的时间和顺序。该学院早期职业发展（Career）项目支持旨在开发自主人工智能系统的研究，该系统可以使软机器人在可验证的安全运动中与人类密切接触，无论如何定义安全。该项目的新颖之处在于首次将施工安全控制应用于软机器人，并专注于其中的独特挑战，如机器人的可控加强，缺乏高度精确的数学模型，以及随着时间的推移机器人的变化。该项目的影响在于有可能降低普通医疗程序的成本，并使其更容易获得，同时减少因临床医生对患者舒适度和疼痛耐受性的主观决定而产生的偏见。为了减少这些偏见，该项目正在将研究与教育结合起来，开展众包公民科学活动。该活动要求高中生构建自己的软机器人，并为自己定义其安全极限，并将他们的结果与他人的结果进行比较，以便我们最终可以回答：什么使软机器人安全，对谁安全？本项目通过寻求三个研究问题的初步答案，将软机器人的控制重新定义为规范满意度问题，而不是运动跟踪问题：（RQ1）软材料的隐含智能是否有助于软机器人机械手满足安全约束，或削弱其这样做的能力，以及在什么条件下？（RQ2）控制策略中的哪些权衡可以平衡软机械臂的固有遵从性，在未知环境中有用，在安全约束的严格性中有用，在环境定义良好时有用？（RQ3）在与人接触不确定的情况下，哪些安全自主合成技术可以表达软机械臂的成功任务转换？为此，该项目进行了相应的创新，有三个研究目标：（RO1）通过控制障碍函数和基于优化的监督控制，为软机械臂开发经过安全验证的反馈系统，满足位置、速度和力的约束。（RO2）将该框架扩展到软机器人的刚度控制，开发一种新的调解/去冲突系统，用于验证多个可能时变软动力学模型下的安全裕度。（RO3）在与人体组织模拟物接触期间，验证软机器人在硬件中安全任务转换的时间逻辑和自动机的使用。总之，该项目为重新定义控制机器人和软机器人之间的关系设定了基线，并有可能协调具身智能和人工智能，从而克服人机接触的基本限制。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Robots built from soft rubbery materials could work closely with humans in settings where rigid robots cannot, since unintentional contact would cause fewer injuries. However, the hope that soft robots are safer around humans neglects how multi-faceted the robot’s tasks may be, and how safety depends on much more than reducing accidental forces. For example, a soft robotic arm assisting in a medical procedure must meet constraints on both too much pressure as well as not enough pressure on a patient’s body, as well as timing and in sequencing of its actions. This Faculty Early Career Development (CAREER) project supports research aimed at developing autonomous artificial intelligence systems that bring a soft robot into close contact with humans during verifiably-safe motion, no matter how safety is defined. The project’s novelty is the first application of safe-by-construction control to soft robots, as well as the focus on unique challenges therein such as controllable stiffening of the robot, lack of highly accurate mathematical models, and changes to the robot over time. This project’s impacts are in the potential to make common medical procedures low-cost and widely accessible, while reducing biases arising from subjective clinician decisions about patient comfort and pain tolerance. To reduce these biases, the project is integrating research with education for a crowdsourced citizen science activity. The activity asks high school students to construct their own soft robot and define its safety limits for themselves, and compare their results with others so that we may ultimately answer: what makes a soft robot safe, and for whom?This project reframes the control of soft robots as a specification satisfaction problem, rather than motion tracking, by seeking initial answers to three research questions: (RQ1) Does the embodied intelligence of soft materials assist a soft robot manipulator in meeting a safety constraint, or detract from its ability to do so, and under what conditions? (RQ2) What tradeoffs in control strategies can balance the inherent compliance of a soft manipulator, useful in unknown environments, with strictness in safety constraints, useful when an environment is well-defined? (RQ3) What synthesis techniques for safe autonomy could express successful task transitions of a soft robot arm during uncertain contact with a human? To do so, the project pursues corresponding innovations with three research objectives: (RO1) Develop safety-verified feedback systems for a soft robotic arm, meeting constraints on position, velocity, and force, via control barrier functions and optimization-based supervisory control. (RO2) Extend this framework for control of the soft robot’s stiffness, developing a new mediation/deconfliction system for verifying safety margins under multiple possibly-time-varying soft dynamics models. (RO3) Validate the use of temporal logic and automata for safe task transitions of a soft robot in hardware, during contact with a human tissue analog. Together, this project sets a baseline for reconceptualizing the relationship between control and soft robots, with the possibility that harmonizing embodied intelligence with artificial intelligence may overcome fundamental limitations in human-robot contact.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.