In nature, animals have adapted a wide variety of approaches to climbing, highly evolved to reliably traverse vertical terrain. Primates use dexterous hands and prehensile tails to ensure strong grasps on tree branches, ibexes plan their foot placement for purchase on rocky cliffs, and jaguars use raw strength to leap over ledges. What each of these mammals have in common is that their motion is controlled by a finely-tuned sensorimotor cortex, requiring accurate perception of the terrain as well as stable control of balance while climbing. These requirements may be prohibitive when building a climbing robot; surface variance degrades gripper reliability, SLAM in highly unstructured environments cannot run in real time on a small processor, and safety margins for strong yet light robots are razor thin. Therefore, we must look elsewhere in nature for realistic inspiration for climbing robots in unstructured terrain.
Cockroaches are able to scale nearly any natural material using microscopic hairs along their legs and feet for surface adhesion. These hairs catch on surface irregularities, known as asperities, and can support the insect’s body through load sharing across many hairs. The cockroach’s scant cognitive intelligence severely limits its ability to plan actions or move dexterously; rather it relies on mechanical intelligence for locomotion. This approach is far more realistic for a field robot that needs to repeatably climb without complex perception systems onboard.
Microspine technology can effectively mimic the adhesion of a cockroach’s foot on a larger scale. Microspine grippers can attach to rough surfaces by sharing the load between hundreds of millimeter scale hooks sprung with hierarchical compliance. Robots such as Kod*Lab’s RiSE , Stanford’s Spinybot , and JPL’s LEMUR have demonstrated the effectiveness of microspines for robotic rock climbing . However, while these systems can scale walls efficiently, their mobility on flat ground is extremely limited by slow moving joints and deliberate foot placement. No legged system has been developed to utilize microspines for climbing while retaining ground agility.
RHex, a cockroach-inspired robotic platform, uses six legs to traverse uneven terrain with a series of open-loop stable gaits. Its leg shape and compliance are key to simplifying control architecture through mechanical intelligence, and has been proven across a wide array of unstructured terrains. We have redesigned RHex’s legs with microspines for rough surface wall climbing. Curved fiberglass RHex legs are characteristically flexible, deflecting under the robot’s weight when in contact with the ground. We have leveraged this compliance with a microspine-tipped design to contribute to the hierarchical compliance required for successful adhesion. By orienting the spines backwards, we ensure that the existing RHex forward gaits are uninterrupted while taking advantage of the microspines for climbing with the typically unused backward walking gait space.
One of the original intents of the RHex platform was for robotic reconnaissance and small payload delivery through rugged environments. Its simple, robust architecture makes it ideal for locomotion through unstructured environments at relatively high speed for a legged system. While much work has been done to allow RHex to survive falls, leap over gaps, climb stairs, and scale obstacles approximately one body length high, RHex’s mobility is largely limited to horizontal terrain  . The platform’s utility and viability when deployed in an urban environment would be improved greatly by adding the ability to scale slopes and walls. This is especially important in areas with incomplete infrastructure such as disaster zones or construction sites.
3 Mechanical Design
The robot described in this paper is an implementation of the standard RHex configuration named “T-RHex,” shown in Fig. 2
. Like RHex, T-RHex is a hexapedal robot with six single degree-of-freedom semicircular legs. Each leg is actuated by a Dynamixel MX-64 servo motor capable of continuousrotation. In contrast to the standard RHex platform, T-RHex is equipped with a “climbing tail.111The tail is mounted to what is the front of the robot during regular walking, which becomes the rear of the robot during climbing.” The tail is actuated with a Dynamixel MX-106 servo motor and can rotate . The fully assembled T-RHex platform weighs 2.5kg and measures 254mm between front and back legs.
3.1 Leg Design
RHex’s legs were redesigned for T-RHex to enable climbing on high-angle terrain. While climbing, T-RHex faces backwards, such that the tips of its appendages are the only points in contact with the surface. A T-RHex leg consists of multiple stacked, thin slices known as a forked wheel leg, as seen in Fig. 3. Each leg slice has a single microspine, and the stacked slices are able to deflect in plane independently of each other, with their shared attachment point concentric with the driving servo horn. The thickness of the slices was driven by microspine size, as each slice had to fully enclose the microspine, leaving only the tip exposed.
Because the microspines protrude only from the tip of each leg, they do not interfere with the ground during normal forward walking. This means that ground mobility is not impacted whatsoever and the robot can travel at the same speed (approximately 1.3 body lengths per second) regardless of whether the legs contain spines or not. This is also useful to prevent surface harm while the robot walks, and to protect the spines from dulling when not in use.
3.1.1 Microspine Design
The microspines on T-RHex consist of size 12 plain shank fish hooks. At approximately 0.6mm thick, these hooks are small enough to allow for 1.5mm thick leg slices. These hooks were chosen for their thickness after preliminary experimentation revealed no effect of hook size on adhesion. Each microspine is attached to the leg slice with a 3D printed PLA piece. These slice “tips” were 3D printed in large batches, and paused mid-print to insert the microspines, which were then printed over. Three different spine angles ( in Fig. 4) were fabricated: , , and . Each spine angle allows for adhesion in a different range of the leg’s rotation, and as such each leg has a mix of spine angles.
3.1.2 Flexure Design
The main portion of each leg slice is constructed of lasercut high-impact acrylic, which is attached to the 3D printed tip with acrylic adhesive on a mechanically interlocking shape.
For effective climbing, the microspines must be able to independently translate while maintaining a fixed angle, due to the narrow range of attack angles for adhesion of microspines. The geometry of the leg slice flexure allows the microspine at the tip to deflect on the order of millimeters both parallel and perpendicular to the climbing surface with minimal rotation of the spine. A full leg stack of flexures has a spring constant of 8-10 kN/m, compared to a typical spring constant of 1.6-1.8 kN/m on RHex legs.
3.1.3 Cassette Design
The full assembly of each leg is a cassette of nine stacked leg slices, with thin spacers in between and rigid outer plates. The rigid outer walls, shown in black in Fig 3(a), prevent out-of-plane bending by the leg slices, but do not come into contact with the surface during walking or climbing. As the legs are spaced slightly farther than one leg length apart, the central legs are spaced outward with standoffs to prevent interference. Each leg is directly bolted to its servo horn.
3.2 Body Design
The body of T-RHex is fabricated from two bent pieces of aluminum sheet metal. The bends in the design serve to stiffen the chassis, keeping it from flexing under load and allowing for thinner sheet metal to be used in its fabrication. This reduces both the weight and the cost of the robot. All of the motors mount to the same chassis, so the leg mount points are rigidly constrained to one another. This also allows for the upper piece, the shell, to be removed when necessary for maintenance. For easy access to the internals of the robot, there is a quickly-removable acrylic lid through which status indicator lights can be seen.
4 Experimental Results
Testing was conducted in several stages to improve climbing performance and validate robot design. Initial trials to determine hook choice, leg shape, spine attachment method, and tine spacing gave clear results later used to justify design choices, the details of which will be here omitted for brevity. The test suites described below were conducted using the completed T-RHex platform with leg shape and body design held constant.
Testing was largely performed on three surface types, chosen to be representative of the various terrains that a small legged robot platform might encounter during deployment in and around a city. The robot’s climbing ability was tested extensively on cork board, brick facade, and plywood. Without a specific deployment goal to dictate target materials, and with limited testing time available preventing an exhaustive study, these surfaces were selected as a minimum subset of surfaces needed to characterize microspine performance on a generalized exploration scenario.
The three surface choices represent the three most common failure modes of microspine climbing feet, as described in . Cork board serves as a soft, pseudo-granular material that microspines can easily puncture and tear through. The expected failure mode is surface degradation due to overloading spines. Plywood is a fairly smooth surface on which microspines must rely equally on existing asperities and active surface deformation by digging into the medium-hardness wood. The failure mode is a lack of strong adhesion points that prevents the microspines from finding purchase. Brick facade serves as a hard, pitted surface that is near ideal for microspines to catch strong, favorably-formed asperities. The failure mode of brick facade is likely tine fracture or plastic deformation of the fish hook when feet do not successfully disengage from unusually strong footholds.
4.1 Static Slope Cling
We performed this suite of tests to determine the effectiveness of the T-RHex leg and microspine design without the added complexity and variability of a custom climbing gait. The leg angle, spine content, and surface material were varied independently while determining the maximum angle at which the robot could statically cling to the wall. The test procedure is as follows:
Set the position of all the robot’s legs to a given angle and command the motors to hold that position indefinitely.
Place the robot at a randomly selected position on the surface while the surface is held horizontal. The robot must be oriented such that the front while climbing will be angled towards the highest point on the surface when raised.
Rotate the test wall about the bottom edge at a rate of approximately 1 degree per second. This rate was determined to be sufficiently quasi-static such that inertial effects of motion could be ignored.
Carefully observe the robot while clinging to note unusual behaviors, interesting microspine adhesion properties, instances of slip-catch motion, and signs of leg damage.
Continue pitching the test wall until the microspines fail and the robot falls. Note the angle of the wall when catastrophic detachment occurs.
The following data was collected across three test surfaces and two leg angles with a sample size of five tests each. As a control, the tests using cassettes of mixed angle microspines were duplicated with a set of blank legs fabricated as tines without embedded microspines. All p-values listed below are the result of one-tailed, heteroscedastic, t-tests.
|Leg Type||Surface Material|
The data indicated that a shallower leg angle of outperformed a steeper leg angle of (), which was critical in later gait design. Performance was similar between cork board and brick, with plywood underperforming by a margin of about
. However, unlike cork board and plywood, which had tightly clustered data points, the performance on brick, a heterogeneous material, was highly dependent on initial robot placement, leading to a much higher standard deviation. This is likely due to the variability of asperity quality between brick face and brick edges. We were able to conclusively () show that the spines improved performance of T-RHex over blank legs for the maximum static cling angle.
We consistently noted that the failure mode of cling was not in the slip of the microspines, but the entire robot body tipping backwards and “peeling” the spines off of the wall as the center of gravity moved away from the wall. Often, the spines on the front and middle legs would be inches away from the wall with the back toes holding all of the robot’s weight after reaching an initial tipping angle of around . This motivated the addition of a tail to T-RHex. We hypothesized that with a smooth tail providing a preload force at a point contact behind and below the robot, the front and middle legs would stay attached to the wall for longer, improving the load sharing between all legs and increasing the maximum static cling angle of the robot.
4.2 Static Slope Cling with Tail
An active tail made of heat-formed acrylic actuated by an already-onboard Dynamixel MX-106 was added to the robot and the same suite of tests was repeated. We assume that the added mass of the acrylic is negligible compared to T-RHex’s total mass and will not affect the results of the testing. The tail was positioned to be in contact with the horizontal surface at the start of the test.
|Leg Type||Surface Material|
The tail succeeded in delaying the robot tip angle, but had mixed results on climbing performance. Because the asperity quality of the plywood was poor, the failure occurred as a slip in what could be modeled as high coefficient Coulombic Friction, largely independent of number of spines engaged. The addition of a tail did not improve the performance of T-RHex on plywood with reliable statistical significance (). On brick, the tail kept more spines engaged to higher angles, which may have improved performance, though with poor statistical significance (. On cork board, the tail actually hurt performance (). We believe that this is due to the mechanism of engagement with a soft surface such as cork board. Our hypothesis is that with more spines in contact there is less preload force per spine, thus they no longer dig in as far, leading to a poorer grip and earlier fall. In all, testing with the tail showed that it is not as beneficial for clinging as we had hoped, but it is extremely reliable as a counterbalance to prevent tipping, which is critical for designing gaits that involve considerable pitching of the robot body.
4.3 Maximum Climb Angle
Limited testing was conducted to determine the maximum angle T-RHex could climb. In these experiments, we demonstrated improvement in climbing ability with the addition of microspines.
Static cling tests were chosen for quantification of climbing ability due to their repeatability, as climbing trials were found to be highly varied. The same test surfaces and fixtures were used as in static cling angle testing. There was slight variation in the microspine leg design between static and climbing tests, with the spines redistributed among legs and rigid sidewalls added after completion of static testing. These modifications did not impact static cling ability, as they served to counter two problems observed during climbing: failure of the microspines to detach and out of plane bending of leg slices during multiple steps. The robot had a tail for all climbing experiments.
Experiments were conducted on the same test surfaces as the static cling tests, with both microspine and blank legs. For each test, the robot was started using the climbing gait described in Section 5 and placed on the climbing surface. The surface was held at a fixed angle, and it was determined whether the robot was capable of making observable (1 cm) upward progress over the course of several steps. Binary search was used to determine the maximum angle the robot could climb for each climbing surface and leg combination, with results compiled in Table 3. Note that microspines greatly outperformed blank legs on all three test surfaces.
|Leg Type||Surface Material|
4.4 Additional Testing
Since the ultimate goal was to climb vertical walls with T-RHex, we spent significant time searching for real-world surfaces on which to test the robot. Our brick facade test surface proved too inconsistent for reliable climbing due to a dearth of asperities on some brick faces. The robot was able to statically cling to vertical surfaces of the following materials: Sycamore tree bark, cinderblock, fabric bulletin board, coarse aggregate concrete, and cubicle wall. The highest climbable slope discovered was on the textured concrete of a building roof.
Though T-RHex was unable to ascend a vertical wall, it surpassed all expectations with “best-case” cling testing. In these trials, performed with the same method as other static cling testing, the robot’s starting position was expertly selected on the brick surface as a point that had particularly intriguing asperities. This testing revealed that T-RHex could reliably hang onto the wall with significant overhang past vertical. In some cases, the maximum cling angle reached , an overhang of .
5 Gait Design
One key aspect of T-RHex is that the robot does not sacrifice the ground mobility of the RHex platform for climbing. Therefore, we decided to keep the RHex’s “alternating tripod” gait for forward ground walking, where the robot maintains at least 3 points of contact on the ground at all times.
We designed custom gaits for the climbing functionality of the robot, inspired by some of the different RHex gaits tested throughout the years. One of the most effective gaits we found was inspired by the RHex stair-climbing gait, where pairs of legs move together to propel the robot upwards. We selected this strategy for T-RHex due to the nature of microspines, which work best when pressed directly at the surface and with symmetric supporting forces on the left and right sides.
We also attempted non-symmetric gaits, such as the backwards alternating tripod, but determined the robot either wouldn’t have sufficient contact with the wall to maintain adhesion, or would pivot about a leg when unopposed by a symmetric climbing force from its mirror.
Ultimately, the selected gait was an adaptation of the “inchworm gait” where the back legs provide propulsion, the front legs engage to hold position between steps, and the tail prevents the back legs from catching or leveraging the body off of the wall when recirculating. Shown in Fig. 10 is a still frame of T-RHex executing this gait. In the frame, we see the back legs are beginning to recirculate, while the front legs engage to prevent the robot from sliding back down. The tail allows the back legs to rotate freely without the danger of contacting the surface.
This gait requires some tuning for different slope angles and surfaces. For instance, steeper angle climbs rely much more heavily on the balance of the tail than shallower angles do. Vertical ascents, however, suffer from what we term as “vaulting,” where the robot attempts to adhere the front toes onto the surface, but instead pushes the robot away from the wall, disengaging the back toes and causing a fall.
6 Future Work
T-RHex is a highly promising prototype, but reaching its full potential will require significant additional work. Future work on refining leg design, enhanced sensing and autonomy, and different modes of control are needed to fully realize autonomous climbing of vertical surfaces.
6.1 Exploring Different Leg Materials
Using acrylic for the T-RHex legs was a decision made for easy prototyping and rapid design refinement. We did not attempt to match the overall spring constant of RHex’s legs, a feature needed for dynamic running. Finding a material that can provide a lower spring constant with equal or greater yield strength will allow for much higher mobility. Additional materials considered include fiberglass, spring steel, and more exotic materials such as metallic glass (which has been used in previous microspine flexures). .
6.2 Increased Autonomy
The onboard software prototyped here only allows for a single gait to be run. For real-world deployment on multi-angled terrains of varying surface material, T-RHex requires increased autonomy the ability for operators to modify gaits without losing power. This comes in two major forms: the floor-to-wall transition and in-situ gait switching.
The floor-to-wall transition will be essential for exploration with T-RHex, but beyond the scope of this paper. The robot should be able to recognize a climbable surface, and automatically position itself at the base of the slope or wall such that the microspines can engage and employ the wall-climbing gait. Achieving this functionality requires scripting a behavior for the transition and adding the ability to use onboard vision or inertial sensing to recognize when to perform the behavior.
6.3 Sensing and Closed Loop Control
As of now, T-RHex moves with an entirely open-loop, scripted gait. We believe that performance can be improved by adding torque sensing to the leg and tail actuators, which would allow the robot to adapt to large-scale surface irregularities in the wall, detect leg slip before catastrophic detachment, and automatically use the tail to balance during wall climbs. This design path would require a platform overhaul, but offers a promising controls-based solution to the shortcomings of our gait design.
We showed that it is possible to augment existing robot architectures with microspines to enable climbing and expand the set of traversable terrain. Our robot, T-RHex, was able to statically hang on surfaces overhanging up to , and climb on surfaces up to , with no impediment to flat ground walking. The T-RHex platform will benefit from exploration of additional leg materials, as well as further gait tuning. Much like the insects that inspired it, this platform has the potential to become a robot that can truly go anywhere.
-  M. Reese. (2017) Cockroach foot. Amateur Micrography. [Online]. Available: http://www.photomacrography.net/forum/userpix/2226_Cockroach_foot_1.jpg
-  M. J. Spenko, G. C. Haynes, J. A. Saunders, M. R. Cutkosky, A. A. Rizzi, R. J. Full, and D. E. Koditschek, “Biologically Inspired Climbing with a Hexapedal Robot,” Journal of Field Robotics, vol. 25, no. 4-5, pp. 223–242, 2008.
-  A. T. Asbeck, S. Kim, W. R. Provancher, M. R. Cutkosky, and M. Lanzetta, “Scaling Hard Vertical Surfaces with Compliant Microspine Arrays,” in Proc., Robotics Science and Systems Conf., Cambridge, MA, June, 2005.
-  A. Parness, N. Abcouwer, C. Fuller, N. Wiltsie, J. Nash, and B. Kennedy, “LEMUR 3: A Limbed Climbing Robot for Extreme Terrain Mobility in Space,” Proceedings - IEEE International Conference on Robotics and Automation, pp. 5467–5473, 2017.
-  A. Johnson and D. Koditschek, “Toward a vocabulary of legged leaping,” Proceedings - IEEE International Conference on Robotics and Automation, 2013.
-  E. Z. Moore, D. Campbell, F. Grimminger, and M. Buehler, “Reliable Stair Climbing in the Simple Hexapod ‘RHex’,” 2002.
-  M. Martone, A. Parness, and A. Willig, “Design and Testing of the Microspine Gripper Tool for the Asteroid Redirect Mission,” Caltech SURF Report.