Introduction

• Living organisms naturally change their shape to adapt to different environments, repair damage, and perform varied tasks.
• Examples include octopuses squeezing through small spaces, caterpillars using peristaltic movement, and salamanders regenerating lost limbs.
• In contrast, most traditional robots are rigid and designed for a single task without the ability to reconfigure their body.
• This research is inspired by biology to create robots that actively change their shape to gain new functionalities and adapt to challenges.
• Dynamic plasticity – the ability to change and adapt physically – is a key differentiator between living systems and artificial machines.

Biological Control of Shape

• Organisms regulate their shape through hierarchical processes, from cellular decisions up to whole-body structure.
• During development, a fertilized egg self-assembles into a precise three-dimensional structure that can adapt if disturbed.
• Regeneration examples include salamanders regrowing limbs and planaria flatworms rebuilding entire bodies from fragments.
• Cells use bioelectric networks to store pattern memories and coordinate shape change even outside the brain.
• This distributed form of intelligence shows how behavior, memory, and physical form are intertwined.

Simulated Shape Changing Robots

• Simulations are used to explore a vast design space since manufacturing multiple robot prototypes is costly and time consuming.
• Evolutionary and learning algorithms help discover nonintuitive designs by testing millions of configurations virtually.
• Simulated robots have learned to change shape to recover from damage, sometimes finding strategies that outperform conventional control methods.
• The design space is enormous even with a few building blocks; simulations help narrow down effective designs before physical realization.
• This process is like trying different recipes in a virtual kitchen until the best one is found.

Physical Shape Changing Robots

• Physical prototypes use multifunctional materials and soft robotics to change shape in the real world.
• Examples include robots that use cable-driven skins to sculpt their inner structure and shape memory alloys to bend or curl their bodies.
• Some designs allow robots to switch locomotion modes, such as changing from a cylindrical rolling shape to a flattened crawling form.
• Techniques like origami folding, inflatable cores, and variable-stiffness materials enable these robots to adapt to obstacles and varied terrain.
• The approach is similar to a sculptor adjusting clay – the robot’s body is reconfigured step by step to suit its task.

Grand Challenges

• There are several major challenges to creating fully adaptive shape changing robots.

Shape Sensing

• Robots need to know their own shape in real time to control their movement and adapt effectively.
• Traditional methods use rigid sensor arrays (like printed circuit boards with accelerometers), but these may not work well on continuously deforming soft robots.
• Emerging techniques include machine learning algorithms and optical fiber sensors to estimate the continuous shape of a robot.
• Designers must develop sensors that can handle stretching, bending, and in-plane strains – much like how human skin senses touch and pressure.

Shape Finding

• Determining the best shape for a robot in a given environment is not straightforward.
• Evolutionary algorithms and simulation can help identify optimal shapes by comparing different configurations.
• Robots must decide when to change shape, weighing energy costs and potential benefits, similar to a chef choosing the best recipe based on available ingredients.
• Current research explores automated pipelines that generate and test many shapes to find the most effective one for tasks such as locomotion or obstacle avoidance.

Shape Changing (Actuation and Control)

• Designing a robot that can continuously morph its structure involves integrating multiple actuation modes (for example, tension, bending, and volumetric expansion).
• Variable stiffness materials allow robots to lock in a shape once it has been achieved, reducing the energy needed to maintain that configuration.
• Control systems must work in closed-loop, constantly adjusting the robot’s shape based on sensor feedback, much like a thermostat regulating room temperature.
• Trade-offs exist between increasing the number of controllable degrees of freedom and the complexity of control and communication among sensors and actuators.

Conclusions and Outlook

• Shape changing robots represent a promising avenue for achieving adaptability similar to that found in biological organisms.
• Bioinspiration offers insights into regeneration, self-repair, and dynamic adaptation that can be applied to robotics.
• Current work in simulation and hardware shows that even small shape changes can lead to significant improvements in functionality.
• Future developments require advances in material science, sensor integration, and automated design algorithms to overcome the remaining challenges.
• Ultimately, these robots could have applications in medicine, search and rescue, and environments where adaptability is critical.

介绍

• 生物体能够自然地改变形状以适应不同环境、修复损伤并完成多样化任务。
• 例如，章鱼可以挤进狭小空间，毛毛虫通过蠕动移动，蝾螈能够再生失去的四肢。
• 与此形成鲜明对比的是，大多数传统机器人都十分刚性，只适合执行单一任务，无法重新构造自己的身体结构。
• 这项研究受到生物学的启发，旨在创造能够主动改变形状、获得新功能并适应挑战的机器人。
• 动态可塑性——即物理上改变和适应的能力——是生物系统与人工系统之间的关键区别。

生物形态控制

• 生物体通过分层过程调控形态，从细胞层面到整体结构均有精确控制。
• 在发育过程中，一个受精卵会自组装成精确的三维结构，即使受到干扰也能进行调整。
• 蝾螈再生四肢、平坦虫从身体片段中重构完整身体等例子说明了再生能力。
• 细胞利用生物电网络储存形态记忆并协调形态变化，这一过程在大脑以外也能进行。
• 这种分布式智能展示了行为、记忆与物理形态之间的紧密联系。

仿真形变机器人

• 由于制造多个机器人原型成本高昂且耗时，研究者利用仿真技术探索庞大的设计空间。
• 进化算法和学习算法帮助在虚拟环境中测试上百万种不同配置，从而发现非直观的设计方案。
• 仿真中的机器人学会通过改变形状来恢复功能，有时这种策略比传统控制方法更有效。
• 即使使用少量构建模块，设计空间也是巨大的；仿真帮助在实现实体机器人之前筛选出有效设计。
• 这一过程类似于在虚拟厨房中尝试各种食谱，直至找到最佳方案。

实体形变机器人

• 实体原型采用多功能材料和软体机器人技术，在真实环境中实现形变。
• 例如，利用缆驱动皮肤来塑造内部结构，以及利用形状记忆合金使机器人弯曲或卷曲。
• 部分设计使机器人能够在不同移动模式间切换，比如由圆柱状滚动变为扁平爬行。
• 采用折纸技术、充气核心和可变刚度材料等方法使机器人能够适应障碍物和多变地形。
• 这种方法类似于雕塑家逐步调整粘土的形状，使机器人一步步重构以适应任务需求。

主要挑战

• 实现完全适应性形变机器人的过程中存在多项重大挑战。

形状感知

• 机器人需要实时了解自身形状，以便有效控制运动和适应环境。
• 传统方法使用刚性传感器阵列（如带有加速度计的印刷电路板），但这些方法可能不适用于持续变形的软体机器人。
• 新兴技术包括利用机器学习算法和光纤传感器来估计机器人的连续形态。
• 设计者需要开发能够处理拉伸、弯曲和面内应变的传感器，就像人体皮肤能感知触觉和压力一样。

形状寻找

• 确定机器人在特定环境下应采取的最佳形状并不容易。
• 利用进化算法和仿真技术可以比较不同配置，从而识别出最优形态。
• 机器人必须判断何时改变形状，同时平衡能量消耗和潜在收益，这类似于厨师根据现有原料选择最佳食谱。
• 当前研究正探索自动化流程，通过生成和测试大量形状来找到在移动或避障等任务中最有效的方案。

形变（驱动与控制）

• 设计能够持续改变结构的机器人需要整合多种驱动模式（例如拉伸、弯曲和体积膨胀）。
• 可变刚度材料使机器人在达到所需形态后能够“锁定”该形态，从而减少维持该状态所需的能量。
• 控制系统必须构成闭环，不断根据传感器反馈调整机器人的形状，就像温控器调节室温一样。
• 增加可控自由度虽然能提升适应能力，但也会带来传感器与执行器之间控制与通讯的复杂性。

结论与展望

• 形变机器人为实现类似生物体适应性的系统提供了有前景的途径。
• 生物启发提供了关于再生、自我修复和动态适应的重要见解，这些都可以应用于机器人设计中。
• 当前在仿真与实体实现方面的进展表明，即使是微小的形态变化也能大幅提升功能性。
• 未来的发展需要在材料科学、传感器整合以及自动化设计算法上取得突破，以克服现存挑战。
• 最终，这类机器人在医学、搜索救援以及需要高度适应性的环境中具有广泛应用前景。