Embodied Intelligence for Advanced Bioinspired Microrobotics: Examples and Insights

TL;DR

The paper addresses the challenge of enabling intelligent behavior in microrobots under severe energy, sensing, and computation constraints. It advocates embodied intelligence (EI) and co-design as a framework to integrate morphology, materials, and environmental interactions into sensing, actuation, and control. Through a portfolio of AMSL robots—Bee++, RoBeetle, SMALLBug, SMARTI, SPARQ, MiniBug, WaterStrider, VLEIBot family, FRISSHBot, and electronics-free soft robots—it demonstrates how EI yields robust locomotion and navigation via physical coupling and reduced reliance on centralized computation. The work argues that EI offers a scalable, robust alternative to classical control for mm-to-cm scale robotics, with broad implications for autonomy in constrained environments.

Abstract

The term embodied intelligence (EI) conveys the notion that body morphology, material properties, interaction with the environment, and control strategies can be purposefully integrated into the process of robotic design to generate intelligent behavior; in particular, locomotion and navigation. In this paper, we discuss EI as a design principle for advanced microrobotics, with a particular focus on co-design -- the simultaneous and interdependent development of physical structure and behavioral function. To illustrate the contrast between EI-inspired systems and traditional architectures that decouple sensing, computation, and actuation, we present and discuss a collection of robots developed by the author and his team at the Autonomous Microrobotic Systems Laboratory (AMSL). These robots exhibit intelligent behavior that emerges from their structural dynamics and the physical interaction between their components and with the environment. Platforms such as the Bee++, RoBeetle, SMALLBug, SMARTI, WaterStrider, VLEIBot+, and FRISSHBot exemplify how feedback loops, decision logics, sensing mechanisms, and smart actuation strategies can be embedded into the physical properties of the robotic system itself. Along these lines, we contend that co-design is not only a method for empirical optimization under constraints, but also an enabler of EI, offering a scalable and robust alternative to classical control for robotics at the mm-to-cm-scale.

Figures (5)

• Figure 1: Embodied intelligence in action. All the microrobots presented in this figure were developed by the author and his team at the AMSL, and each exhibits some form of EI at some level in its design. (a) The $95$-mg Bee++, which leverages FSI and an ISP-based mechanism in combination with modern nonlinear Lyapunov-based methods, to control its yaw DOF. (b) The $88$-mg RoBeetle, an autonomous anisotropic-friction-based crawler mechanically powered by an NiTi--Pt composite catalytic artificial muscle. This muscle uses the flameless catalytic combustion of methanol, enabled by a rough layer of Pt, to thermally excite its core made of NiTi SMA in cycles that produce periodic actuation. The system that controls the catalytic-combustion process and the phase transitions of the SMA material is entirely electronics-free, thus leveraging EI.~(c) The SMALLBug, a crawler that uses a high-frequency SMA-based bending actuator to locomote on flat surfaces. In this design, the cyclic bending motion of the driving actuator is transformed by a $2\Sigma$-shaped frame into rectilinear locomotion through leveraging anisotropic friction.~(d) The SMARTI, a crawler composed of two SMALLBug platforms connected in parallel. This configuration is $2$D steerable simply by exciting its two driving actuators with phase-shifted PWM voltages, thus leveraging anisotropic friction for functionality and control.~(e) The SPARQ, a fully autonomous---from both the power and control perspectives---crawler, which was conceived as an advanced version of the SMARTI. (f) The MiniBug, which, at 10 mg, is the smallest and lightest crawler with onboard actuation ever created. This robot was developed by miniaturizing the actuator and frame of the SMALLBug, and by adopting the legs of the SPARQ. (g) The WaterStrider, a surface swimmer that uses rowing-resembling motion patterns of two paddles and anisotropic drag to propel itself forward. (h) The VLEIBot (left) and VLEIBot+ (right) are two robots that use propulsors inspired by anguilliforms to swim forward and steer themselves. These propulsors generate thrust by leveraging FSI as traveling waves are induced by flapping soft passive fins with SMA-based bending actuators. (i) The VLEIBot++, a fully autonomous---from both the power and control perspectives---swimmer, which was conceived as an advanced version of the VLEIBot+. (j) The BILLEBot, a swimmer design that combines the transmission mechanism of the WaterStrider with the traveling-wave-based thrust-generation method of the VLEIBot platform. (k) The old FRISSHBot (left) and new FRISSHBot (right) are two swimmers composed of two plates connected by an SMA-based bending actuator that applies periodic torques---with equal magnitudes and opposite directions---to both of them during operation. These actuation torques induce hydrodynamic reactive torques---generated by aggregated inertial and viscous forces---on the plates that, as a consequence, produce the thrust required for swimming. Specifically, by design, the front plate functions as an anchor and the rear plate as a caudal flapping rigid fin. This flapping fin, through the creation of a couple of vortices during an operation cycle---one clockwise and the other counterclockwise---induces a jet that, by conservation of momentum, propels the robot forward. (l) The pneumatic soft robot on the left can locomote inside pipes and trenches by leveraging pressure-controlled anisotropic friction, using a distributed stretchable artificial skin as the main sensor. The pneumatic soft robot on the right can locomote inside pipes and trenches driven by an entirely electronics-free feedback controller based on neuromorphic mechanical computation, using mechanical tactile and proprioceptive sensors.
• Figure 2: The Bee++ and its flapping modes.(a) The $95$-mg Bee++ is the first four-actuator four-wing robotic insect ever developed. As discussed in BenaRM2023, this flyer is six-DOF controllable during flight. (b) Each wing is connected to its respective unimorph actuator through a four-bar transmission mechanism installed with an inclination $\beta$ rad relative to the $\boldsymbol{b}_2$-$\boldsymbol{b}_3$ plane---equivalent to an inclination $\left(\pi/2 - \beta\right)$ rad relative to the $\boldsymbol{b}_1$-$\boldsymbol{b}_2$ plane. (c) The $\beta$-inclined installation of each transmission of the robot is the result of a feature of the actuator design, whose connecting element at its distal end is tilted by the angle $\beta$. (d) The Bee++ is underactuated as it has fewer actuators than degrees of freedom; however, it can be controlled using four basic flapping modes and nonlinear Lyapunov methods. The first mode consists of flapping the four wings of the robot with identical frequencies and amplitudes; thrust can be modulated by varying the amplitude of flapping during flight. The second mode consists of flapping the wings of the robot in pairs and asymmetrically with respect to the $\boldsymbol{b}_1$-$\boldsymbol{b}_3$ plane; the roll DOF can be modulated by increasing---or decreasing---the amplitudes of flapping of one pair of wings relative to those of the other pair. The third mode consists of flapping the wings of the robot in pairs and asymmetrically with respect to the $\boldsymbol{b}_2$-$\boldsymbol{b}_3$ plane; the pitch DOF can be modulated by increasing---or decreasing---the amplitudes of flapping of one pair of wings relative to those of the other pair. The fourth mode consists of flapping the wings of the robot in diagonal pairs; the yaw DOF can be modulated by increasing---or decreasing---the amplitudes of flapping of one pair of wings relative to those of the other pair. Note that the inclination of the stroke plane relative to the $\boldsymbol{b}_1$-$\boldsymbol{b}_2$ plane generates a yaw torque that can be modulated to actuate and control the yaw DOF, thus exemplifying EI in action.
• Figure 4: Functionality of the catalytic artificial muscle developed to drive the RoBeetle shown in Fig. \ref{['FIG01']}(b).(a) During a full actuation cycle, the material of an SMA wire transitions between three distinct crystal-structure states: detwinned martensite, austenite, and twinned martensite by the sequential application of heat, allowance for convective cooling, and application of stress. (b) In the case of the RoBeetle's muscle, controlled heat is applied through the catalytic combustion of methanol---facilitated by a rough layer of Pt, a multipurpose catalyst---and stress is applied using a CF leaf spring (shown in Fig. \ref{['FIG05']}(b)). Accordingly, an actuation cycle starts at the detwinned martensite state---corresponding to the extended state of the wire at room temperature; then, after surpassing the transition temperature of the SMA material, the system transitions to the austenite state---corresponding to the contracted state of the wire; next, after cooling down through free convection and reaching room temperature, the system transitions to the martensite state. The stress applied by the leaf spring to the wire continually detwins the SMA material.~(c)--(e) SEM images of the surface of an NiTi-Pt composite wire with a diameter of $87$ µm. The rough and porous catalytic layer (Pt-black) has a thickness of $18.1$ µm. The magnifications of the images are ${\times}350$, ${\times}1\space200$, and ${\times}6\space500$; the scale bars indicate distances of $100$, $30$, and $5$ µm, respectively.
• Figure 6: Mechanisms of locomotion, leveraging EI, of some of the robots discussed in this article.(a) Bending actuator (left) and anisotropic-friction-based locomotion mode of the SMALLBug platform shown in Fig. \ref{['FIG01']}(c).~(b) Swimming thrust generated by a traveling wave passing through the unactuated soft slender tail of a VLEIBot prototype shown in Fig. \ref{['FIG01']}(h).~(c) Swimming thrust generated by tail-generated vortices used by the FRISSHBot platforms, shown in Fig. \ref{['FIG01']}(k), to propel themselves forward.~(d) Basic locomotion mechanism used by the pneumatic soft robots shown in Fig. \ref{['FIG01']}(l).
• Figure 7: Photographic composites of images showing locomotion tests of six of the robots discussed in this article.(a) A RoBeetle prototype crawling on a flat surface. (b) A MiniBug prototype crawling on a flat surface, excited with a $15$-Hz PWM signal with a DC of $10$ %.~(c) A WaterStrider prototype swimming excited by four different PWM signals with the frequencies and DC values in parentheses. (d) A BILLEBot prototype swimming excited by two different PWM signals with the frequencies and DC values in parentheses. (e) A FRISSHBot prototype swimming excited by two different PWM signals with the frequencies and DC values in parentheses. (f) Electronics-free pneumatic-based soft robot, shown on the right in Fig. \ref{['FIG01']}(l), crawling inside a trench. This robot is feedback-controlled using neuromorphic mechanical computation and mechanical tactile and proprioceptive sensors.