A Hexapod Robot Modeled on the Stick Insect, Carausius morosus William A. Lewinger, Member, IEEE , H. Martin Reekie, Member, IEEE , and Barbara Webb 

Abstract— Robot builders have often used insects as a source of inspiration when designing their mechanical systems, due to their ability to easily navigate uneven terrain, overcome or avoid obstacles, and adjust gaits based on traveling speed. Robotics has borrowed from nature with varying degrees of abstraction, from physical appearance to observed behaviours. This paper describes the design and construction of a robotic hexapod based on the stick insect, Carausius morosus. Physically, it is an 18.8:1 scale representation of the insect with 3-DoF legs. The to-scale design was chosen to provide similar physical attributes, such as joint and leg locations, sizes, and ranges-of-motion, which will allow more meaningful comparisons between robot performance and actual insect movements (as opposed to arbitrary hexapod designs). A custom-designed leg control board is responsible for deciding leg joint movements based on a model of the neurobiological systems identified in the insect. A distributed network of six boards will be used to control the legs based on internal parameters that can be modulated by descending commands or adaptively altered by ascending sensory signals when interacting with the environment. Our final aim in this work is to add a vision system to create depth maps, which will be used as an input to a learning system, coupled with the mechanical sensory system, such that terrain that triggers reflex actions can be associated with visual cues in order to predictively avoid obstacles and potholes.

I

I.

INTRODUCTION

sensory systems have been incorporated that allow the software to actively change its actions such that irregular terrain can be navigated. While adaptive to uneven terrain, or even capable of implementing reflex behaviours that allow the robots to overcome obstacles and navigate somewhat sparse terrain, these robots mostly use engineering methods based on observed insect actions to achieve these goals, rather than directly emulating the underlying neurobiological systems. In recent years, insect neurobiological systems have been identified that are responsible for controlling the movements and orchestration of leg joints such that coordinated actions lead to the generation of stepping motions [9]. The discovery of the neural pathways represents the first opportunity to implement the underlying control of leg movements, not just externally-observed motions. The actions of these pathways have been modeled for use in simulations [5], with single- and two-leg test platforms [11], and a complete physical hexapod (BILLAnt-a) [10]. However this hexapod was not a to-scale representation of any insect. As such, stepping motions needed to be modified from those seen in nature, due to physical range-of-motion limitations of the robot’s mechanical design.

NSECTS are often used for inspiration when designing walking robots, in part for their ability to easily navigate irregular terrain. Their six multi-jointed legs are highly adaptive, which allow them to readily change their stepping patterns to match the terrain. Sensory systems within the legs allow this adaptation to occur in a rapid and distributed manner, while maintaining overall co-ordination. Previous hexapod designs, such as the Tarry series [7], Robot II [6], and the LAURON series [8] have based their leg control and walking systems on observations of insect behaviour. These observations have been programmed into the control software to roughly emulate the movements of insect leg segments during stepping actions. Additionally,  Manuscript received March 19, 2011. This work was supported in part by the Engineering and Physical Sciences Research Council (EPSRC) under Grant EP/E063322/1. W. A. Lewinger is with the University of Edinburgh School of Informatics, Institute for Perception, Action and Behaviour, Edinburgh, EH8 9AB, UK. e-mail: [email protected]. H. M. Reekie is with the University of Edinburgh School of Informatics, Institute for Perception, Action and Behaviour, Edinburgh, EH8 9AB, UK. e-mail: [email protected]. B. Webb is with the University of Edinburgh School of Informatics, Institute for Perception, Action and Behaviour, Edinburgh, EH8 9AB, UK. e-mail: [email protected].

Fig. 1. Design rendering of BILL-Stick with the thorax shell (top). Assembled BILL-Stick mechanical system: skeleton with thorax shell support spines and legs with servo motor joint actuators (bottom).

The work presented here, which is part of the Neuromorphic Sensorimotor Integration for Legged

Locomotion (NSILL) project, represents a new hexapod design, the Biologically-Inspired Legged Locomotion Stick insect (BILL-Stick), which is based more closely on an actual insect, Carausius morosus. The robot’s dimensions have been taken directly from the insect in order to create an 18.8:1 scale model (Fig. 1). The single leg control system is modeled on the insect’s neurobiological system, while the gait generation method is based on externally-observed stepping patterns, as the underlying neurobiological systems have not yet been identified. In addition to creating a to-scale model of a biological system using neurobiological methods of leg control, the BILL-Stick hexapod is also designed to incorporate vision and learning systems, implemented in neuromorphic hardware, that will allow the robot to recognize visual scenarios where, previously, obstacles have triggered reflex actions. These systems will enable the robot to pro-actively alter its stepping motions in order to navigate past such obstructions without requiring an actual collision to trigger a reflex action. Custom-designed leg controller boards have been developed for this project that surpass many small, off-theshelf microcontroller solutions currently available in terms of control loop speed. The distributed processing network to both control individual legs and the generation of stable gait patterns is able to operate at speeds up to 1kHz. At this speed, the controllers are able to fetch control parameters from an on-robot computer, execute the neurobiological leg control model, and return the leg-generated data to the computer. These controllers are also capable of interfacing with a variety of motor types and sensors, such as standard hobby servos, PWM-controlled dc motors with position feedback, and I2C-based devices. In this paper we describe in detail the mechanical design of the hexapod, the electronic design of the controller boards and the overall system architecture. Results for walking and learning will be covered in a future publication. II. MECHANICAL DESIGN A. Physical Aspects The physical dimensions of the robot are based on insect data collected from Cruse [1] and Pfeiffer et al. [13]. These publications presented body and leg segment lengths, joint attachment angles, and joint ranges of motion. The dimension data were scaled (Table I) and then used to create the hexapod. Angular data were used without modification (Table II). The skeleton (Fig. 1, bottom) is used as a central mounting device for the six legs, and also supports the control systems, such as the two Central Computers, the leg control network, and on-board Li-Ion batteries. The central skeleton is constructed from 1.27cm (0.5in) O.D., 1.02cm (0.402in) I.D. 6061 aluminium tubing. Tubing was chosen as a construction medium for its high strength and low mass

when compared to solid structures. The tubing sections connect the leg pairs and extend to the head and tip of the abdomen. Connection nodes are comprised of a 2.54cm (1.0in) square aluminium cube and tubing mounting hubs acquired from the Lynxmotion Servo Erector Set (SES) (Lynxmotion, Inc., Pekin, IL, USA). The tubing parts used here have slightly thicker walls than those available from Lynxmotion, which provide the greater strength necessary for the large size of the hexapod, and so the tubing mounting hubs were turned down on a lathe to accommodate the smaller inner diameter. Each of the leg pairs (Fig. 2) is attached to the skeleton through a thoracal node that was custom-designed to create the appropriate α and β angles for each pair. These parts serve as attachment points for Lynxmotion SES servo mounting brackets, which hold the thorax-coxa (ThC) joint actuators. These actuators are responsible for protraction and retraction of the leg.

Fig. 2 Right-front leg of the robot with labeled features.

The ThC joint is fixed to the coxa-trochanter (CTr) joint by means of custom-designed brackets. These two-piece brackets wrap about the mid-sections each of the joint actuators. The actuators are secured to the brackets by mounting screws. The two joint brackets/actuators are attached to one another by screws. This rigid assembly represents the coxa. The distance from the ThC joint axis of rotation and the CTr joint axis is 2.82cm (1.11in). This distance is the smallest, indivisible length in the robot, as it is determined by the size of the motor, and is responsible for dictating the 18.8:1 scale with the insect. The CTr joint is attached to the femur-tibia (FTi) joint by an SES servo bracket attached to an aluminium tubing segment representing the femur. The tubing (with the same diameter dimensions as the skeleton) is affixed to the FTi joint by a modified SES tubing mounting hub and another custom-designed bracket that surrounds the FTi joint actuator.

TABLE I PHYSICAL DIMENSIONS

Section

Length

Skeleton Head to Abdomen Head to Prothorax Head to Mesothorax Head to Metathorax Prothorax Legs Coxa Femur Tibia Mesothorax Legs Coxa Femur Tibia Metathorax Legs Coxa Femur Tibia Thorax Shell Support Spines One at each thorax section, and one each at the head and abdomen tip Thorax Outer Shell Length Diameter

139.59cm 13.35cm 46.24cm 66.73cm 2.82cm 27.63cm 25.75cm 2.82cm 21.43cm 20.11cm 2.82cm 23.12cm 22.18cm 7.38cm 139.59cm 10.16cm

TABLE II JOINT ANGLES

Section Prothorax Legs α (measured from the sagittal plane) β (measured from the horizontal plane) ThC RoM (measured from the sagittal plane) CTr RoM (measured from the horizontal plane) FTi RoM (measured from an extension of the femur) Mesothorax Legs α (measured from the sagittal plane) β (measured from the horizontal plane) ThC RoM (measured from the sagittal plane) CTr RoM (measured from the horizontal plane) FTi RoM (measured from an extension of the femur) Metathorax Legs α (measured from the sagittal plane) β (measured from the horizontal plane) ThC RoM (measured from the sagittal plane) CTr RoM (measured from the horizontal plane) FTi RoM (measured from an extension of the femur)

Angle (deg) 75 -40 +5 – +155 -90 – +90 -180 – 0 85 -40 +15 – +165 -90 – +90 -180 – 0 135 -40 +65 – +215 -80 – +100 -180 – 0

Two lift-assist springs are attached between the femur servo bracket and the CTr joint actuator bracket. These 4.45cm (1.75in)-long extension springs have a k value of 535.7g/cm (3.0lbs/in) and are used to aid the CTr joints in support of the robot’s central body mass. The role of these springs is crucial as the CTr joints have the highest amount

of load to support when walking, which would greatly reduce their lifespan without the springs. The tibia is comprised of another SES servo bracket and tubing assembly, which is connected to the output of the FTi joint actuator. The distal end of the tibia is covered with a rubber cap to provide some traction when walking on smooth surfaces, and also to eliminate damage to the tibia end and substrate when walking. There is no tarsus section for this robot. At each of the three thoracal nodes, and at the head and tip of the abdomen, thorax shell support spines are attached to the top sides of the 2.54cm (1.0in) SES cubes. These 7.38cm (2.91in)-long spines are assembled from standard (unmodified) tubing mounting hubs, 3.81cm (1.5in)-long SES aluminium tubes, and an upper 2.54cm (1.0in) aluminium cube. The upper cubes are used as mounting points for brackets that hold the leg controller electronics and Li-Ion batteries. Tubing mounting hubs fit through mounting holes in the dorsal side of the thorax shell to hold it in place. Portions of the ventral side of the thorax are cut away to allow the shell to be placed over the skeleton and the internal electronics. B. Joint Actuators The leg joints are actuated by modified hobby servo motors, the Hitec HSR-5990TG (Hitec RCD, LLC, Poway, CA, USA). These motors were chosen for their high torque (stall torque: 2.94Nm, 416.6oz-in), titanium-geared transmission, and their ability to operate up to 7.4vdc, which allows them to be powered by 2-cell Li-Ion batteries without the need for voltage regulation. Because of their standard hobby servo dimensions, these motors are able to be used with the Lynxmotion Servo Erector Set brackets, which minimized the number of custom-designed parts that were required. Each of the three joint actuators per leg directly drives its associated leg segment. By attaching the leg segments directly to the servo output horns, the mechanical design of the joints is simplified. This direct connection also allows the joints to take advantage of the nominal ±90deg range-ofmotion (RoM) of the servos, which is approximately the same RoM as the insect (Table II). It is important for the leg control system to know current joint angles (servo position) and joint loads (current consumption). As this information is not available from standard hobby servos, the motors were modified. The servo’s internal PCB, which is responsible for receiving PWM position commands from a host and converting those commands into servo output positions, has been removed, leaving the motor, transmission, and output potentiometer in the servo case. The two motor leads and three potentiometer leads are connected to a channel on the Motor Driver Board and Controller Board (described below). These boards execute a closed-loop PID controller, while monitoring servo position and current consumption.

III. LEG CONTROLLER DESIGN Each leg has its own, custom-designed controller responsible for controlling the joint actuators. These distributed controllers command the joints of a single leg to perform stepping motions, while adapting those motions based on position and load information sensed from the joints. The leg controllers are also connected together in a network for data sharing (needed for gait generation) and data routing (used to deliver and collect data from the central computer). Three types of PCB are used in the robot: Controller Board, Motor Driver Board, and an Interconnect Board. A. Controller Board This board is the key component for the leg control system (Fig. 3). It is 85mm x 42mm (3.35in x 1.65in) in size and contains one PIC18F4550 and four PIC16F616 microcontrollers (Microchip Technology, Inc., Chandler, AZ, USA). The PIC18F4550 acts as the master for the board, and the four PIC16F616s, one for each motor channel, are the slave units, which are capable of operating independently from one another.

The Controller Boards are capable of operating in several modes: connected to a PC via an Interconnect Board and a Router Board; connected directly to a PC; or as a standalone unit. For testing purposes, the board is connected directly to a PC via USB and uses a custom interface. This interface allows the user to actuate individual joints on a single leg and receive leg data such as position and load. When implemented in the robot, a network of Controller Boards is created where the boards are coupled to one another via an Interconnect Board. The Router Board is also part of the network through the Interconnect Board (Fig. 4). Board-to-board communications occurs by means of a custom interface which has some similarities to RS232, but it is voltage-driven and has eight channels operating simultaneously, such that byte-wise, rather than bit-wise, data transfers can take place. The Router Board is used to connect to the leg control network to the central computer through a USB cable. On each leg controller, the Master PIC is responsible for using its walking parameters to generate stepping motions. Normally, these parameters are sent by the central computer; however, they could also be pre-programmed so that the controller can operate without a host.

Fig. 4. Leg control system network. Each leg has an associated Controller Board and an attached Motor Driver Board that are responsible for creating stepping motions for its associated leg. The six legs are connected to one another via an Interconnect Board through an 8-parallelchannel bus. This bus also connects the leg controllers to the Router Board, which is used to interface with the Host Computer via USB.

Fig. 3. Controller Board with attached Motor Driver Board (top). Block diagram of the man Controller Board and Motor Driver board sections (bottom).

Input voltage can range from 5V dc to 12V dc. This voltage is regulated to 5V dc for the PIC microcontrollers, motor control PWM signals and the I2C bus. Raw input voltage is passed to the attached Motor Driver Board for motor power.

The Master PIC executes its leg control software, which has been hand-programmed in assembler, to optimize performance and maintain absolute control over system timing. The Slave PICs interface with the joint actuators by a PID loop and have also been hand-programmed in assembler. Computation cycles use the USB frame as a “heart-beat” for the system. Six asynchronous interrupt transfers to separate endpoints on the same PIC allow 256 bytes of information to be sent from the central computer to the

Router Board, and 128 bytes to be received back to the central computer, in each 1ms USB frame. Use of interrupt transfers ensures high-priority on the USB bus; more conventional bulk transfers are given the lowest priority on a USB bus. On reception of walking parameters from the central computer, the Router Board transmits these parameters to each of the Controller Master PICs. Each Master PIC then sends a packet of previous cycle leg data (joint positions, loads, current states, etc.) to every other master and also to the Router, which is then sent back to the central computer for data collection and storage, and to be used by other subsystems within the robot. Having received walking parameter data, the Master PIC performs its calculations to determine new joint positions, and sends those commands to the Slave PICs. This occurs during the 1ms USB frame time. The Slave PICs then actuate their respective joints through a PID loop, while monitoring motor current consumption for an “over-current” situation. If an over-current level is reached (typically set at 5A per channel), the individual motor channel is disabled until the situation is resolved. Position and current feedback is received both by the Slave PIC, in order to perform the PID loop, and the Master PIC. The Master PIC uses this feedback to determine its next set of joint position commands, as defined by the leg control software. The USB frame timing scheme means that information flows from the central computer to each of 24 slaves within a period of 1ms, but there can be a delay of up to 2ms for information to come back from a Master PIC to the central computer. This delay is caused by the system requiring two computation cycles to transfer data from the Master PICs. While this is the operating mode of the robot, the Controller Boards are also capable of driving standard hobby servos with a hobby PWM signal, or communicating with other servos or devices via I2C or SPI. The additional communication modes, coupled with the variety of controller implementation options (stand-alone, PCconnected, PC-router-connected) makes this a powerful, flexible controller for future projects. B. Motor Driver Board In order to directly drive dc motors, such as the modified servo joint actuators, a Motor Driver Board is attached to each of the six Controller Boards. Its edge can be seen under the Controller Board shown in the top of Fig. 3. The Motor Driver Board is 78mm x 41mm (3.07in x 1.61in) in size and contains four, independent H-bridge circuits, one for each motor channel (an associated Slave PIC) of the Controller Board. These H-bridges are capable of supplying over 5A per channel. The two boards connect directly to one another through two rows of header pins and sockets. High-speed PWM signals and direction bits from the

Slave PICs are inputs for the H-bridge circuits. These circuits drive the joint actuator motors. The joint position signals are passed from the Motor Driver Board to the Slave PICs for use in the PID control loops. The current consumption of each motor is also passed to its respective Slave PIC. C. Router Board The Router Board is a Controller Board without an attached Motor Driver Board. There is no physical difference between the Router and Controller, only its role is different. While the six Controllers communicate with one another, and the Router Board, via a custom bus, the Router Board uses a USB connection to transfer data to/from the central computer. As with the Controller Boards, the Router holds walking parameters and leg data for all six legs. Walking parameters received from the central computer are distributed to the Controllers, and data collected from the legs is sent to the central computer. These transfers occur during each computation cycle. D. Interconnect Board The Interconnect board is used to couple the custom buses between the six Controller Boards and the Router. Data transmission and board enable lines from the 2x8 connectors on the other boards converge here through some latch and logic circuits. When a single leg is tested, with the Controller Board either in direct connection to the central computer, or through the Router Board, the Interconnect Board is not required. E. LCD Board When using a single Controller Board connected to the central computer for testing, an LCD Board can be attached to the 2x8 connector in place of, or in addition to, the Interconnect or the Router Board. The LCD Board is used for diagnostic purposes and during initial setup and configuration of the leg control parameters. While it is an essential tool, it is not part of the leg control network when operating the complete robot. IV. WALKING SYSTEM Control of the legs in an organized, yet adaptive manner is the role of the walking system. The walking system consists of the central computer to provide walking parameters, the leg controllers to execute the leg control and gait generation algorithms, and the legs themselves. Data generated by the legs, such as joint positions, loads, and states is stored in the central computer database. A. Leg Control Building on previous work ([5], [10], and [11]), the leg control method employs two systems: a pattern generator to

generate nominal stepping motions; and a reactive system that can alter the nominal motions through interactions between the robot and the environment. Each of the three joints in each of the six legs has its own two-state oscillator capable of generating rhythmic patterns. The joints within a single leg share a common period for their oscillators, but vary in duty cycle and phase shift. These oscillators control the two directions that each joint can move, and have their parameters designed to create regular stepping motions, suitable for navigating smooth, level terrain. The pattern generator, which uses only internal cues, can also actuate joint motors in the absence of externally-generated feedback signals such as the joint’s actual position or its load. When traversing irregular terrain, the reactive portion of the leg control system overrides the pattern generators (Fig. 5). This portion uses a rule-set model of the neurobiological system found in the insect [9]. Each of the three leg joints remains in one of its two states (directions of motion) until sensory information dictates that a state change should occur. This allows the stepping motions to be reactive to elevation changes, and obstacles, which creates a robustly adaptive manner of walking.

should have occurred. When ground contact is not detected, the joint states are temporarily altered to retraction and levation. This allows the leg to rise quickly in order to step out of and beyond a hole or gap. Once the reflex-triggering event has past, normal walking resumes. This process can repeat until leg re-discovers the substrate. When a leg collides with an obstacle taller than its nominal step height, the elevator reflex is triggered. This reflex has a similar reaction to the searching reflex in that the joint states are temporarily set to retraction and levation. The retraction command clears the leg from contact with the obstacle and the levation action is used to raise the leg above its usual step height in order to clear the obstacle. After the temporary period, the normal pattern generator and reactive stepping methods resume. If the obstacle is not cleared in a single step, the process repeats. C. Gait Control Observations of insects have yielded a set of rules to create coordinated gait patterns [2], [3]. An adaptation of these rules, which has been incorporated in the reactive portion of the leg control method, uses signals from adjacent legs in order to generate stable, speed-dependent gaits [12]. The ThC joint angles are used as representatives of how far a leg has progressed along its step path. These signals are used to influence the stepping patterns of neighboring legs. The gait generation method is based on external observations of walking insects, as the underlying neurobiological systems have yet to be identified. As implemented in previous work, this method is computationally simple, and is performed by the leg controllers without affecting their high speed of operation. V. CONTROL INFRASTRUCTURE

Fig. 5. Sensory-Coupled Action Switching Modules (SCASM) leg control method. This method is modeled after the neurobiological systems found in Carausius morosus, and uses bi-stable oscillators that are altered based on sensory cues from joints within the leg. These sensory cues are generated by interactions with the environment allowing the system to be robustly adaptive to irregular terrain.

B. Reflexes Embedded within the reactive portion of the leg control system are biologically-plausible sensorimotor pathways that provide searching and elevator reflexes. These reflexes are used to navigate past obstacles that would otherwise hinder the robot. The onset of each reflex is stored in the central computer database for use by the Learning System. The searching reflex is evoked when a hole or gap is encountered. It is triggered when the CTr joint depresses below a threshold joint angle by which ground contact

There are two central computers that represent the control infrastructure (Fig. 6). Each computer is a fit-PC2 (CompuLab, Haifa, Israel) with a 1.6GHz Intel Atom Z530 processor, 1GB of RAM, and a 160GB HDD. Ubuntu 10.10 is installed as the operating system on each device. The first unit, Central Computer #1, is used to interface with the leg control network (specifically, the Router Board) via USB. Walking parameter data, which is stored in a database located on the HDD is sent to the leg control network. Data generated by the leg control network is then received and stored in the same database. A 3-D physics engine simulator also resides on Central Computer #1. This simulator interfaces with the database similarly to the leg control network, in that it reads walking parameter data, performs simulated walking calculations, and returns its generated leg data. The simulator can be used in lieu of actuating the physical robot. Additionally, it can be used to test new or replay previous walking parameters without the robot. Central Computer #2 is connected to Computer #1 via a peer-to-peer Ethernet cable to create a two node network.

Its role is to interface with the onboard stereo vision system (Surveyor Stereo Vision System, Surveyor Corporation, San Luis Obispo, CA, USA) in order to generate depth maps as the robot is moving. These depth maps act as inputs to the Learning System.

Stepping results shown in Fig. 7 show similarities to previously observed insect foot paths. Differences between the two sets of plots are due to differences in the insect muscle behaviour and that of the artificial “muscle model” used in the leg control software. Alterations to this model are still in progress to provide more consistency between the robot and animal performances.

Fig. 6. Overall system block diagram. Two on-board computers are used to interface with the walking system and camera peripherals. Central Computer #1 passes walking parameters to the leg control network and receives data generated by the legs. It also generates a simulation of the robot system. Central Computer #2 is used to interface with the stereo vision system and perform the learning algorithm, which learns to correlate reflex actions when encountering obstacles with visual field depth information.

The Learning System also receives input from the database located on Central Computer #1. This input informs the Learning System when the robot encounters an obstacle that generates a reflex action in an individual leg. When such actions occur, the Learning System begins to associate visual depth information with reflex triggeringobstacles. As the association strengthens, the Learning System will alter the walking parameters stored in the database such that the leg control network will preemptively navigate past obstacles without engaging reflex actions. VI. INITIAL STEPPING RESULTS To perform initial tests necessary to evaluate the leg controller, stepping control software, and allow the definition of configuration parameters for the stepping controller, a stand was constructed to support a single leg of the hexapod. The stand allowed full movement of the leg, while providing the ability to adjust static conditions such as ThC joint height. During this testing, the right-middle leg was used and its three joints were controlled by a single control module, consisting of a Controller Board, a Motor Driver Board, and an LCD Board. The controller was connected to a host PC that executed the stepping control software and recorded data generated by the leg. The PC-based stepping software, programmed in C, was used to allow for maximum flexibility in stepping routines and configuration parameters, without the burden of constantly re-programming the Master PIC on the Controller Board. The C language was chosen as this will be the basis for the compiled assembly code that will ultimately reside within the Master PIC.

Fig. 7. (top) Stick insect right-middle leg foot path plots in the xz- and xyplane (re-created from [4]). (bottom) xz- and xy-plane foot path plots for the right-middle leg of the robot. The lower plots also indicate the joint states for each of the three DoFs during the single step.

VII. CONCLUSIONS The work presented here describes a hexapod robot based on the physical and neurobiological aspects of the stick insect, Carausius morosus. It has been designed to

implement leg control methods as identified in the insect neurobiology and gait generation methods based on insect observations, and combine these with a learning system that is capable of identifying and predicting reflex-triggering events. By building a to-scale replica of the joint and leg locations, sizes, and ranges-of-motion, more meaningful comparisons between robot performance and actual insect movements will be possible. Control of the leg actuators, and execution of the leg control and gait generation models, is performed by a network of custom-designed controller boards. These boards allow control loop speeds of up to 1kHz, which provide the robot the means to navigate its environment smoothly. The program speed and flexibility of these controllers make them ideal for this project and future research. The hexapod has been designed to navigate smooth and irregular terrain, navigate reactively past obstacles, and learn to associate visual cues with obstacles such that they can be pre-emptively cleared. Upcoming tests will quantify the performance of the robot during these tasks. ACKNOWLEDGMENTS The authors would like to thank Rich Bachmann for CNC-ing the custom-designed parts for the robot, and the members of the Mechanical and Electrical Workshops at the University of Edinburgh, Hugh Cameron, Douglas Howie, and Robert MacGregor, for their valuable assistance. REFERENCES [1] [2] [3] [4] [5] [6]

[7]

[8]

[9]

H. Cruse. (1976) “On the Function of the Legs in the Free Walking Stick Insect Carausius morosus,” in J. Comp. Physiol., vol. 112, pp. 235–262, 1976. H. Cruse, J. Dean, U. Müller, and J. Schmitz. (1991) “The Stick Insect as a Walking Robot,” in Proceedings of the Fifth Int. Conf. on Advanced Robotics. Vol. 2 , S. 936-940 – ICAR, 1991. H. Cruse, T. Kindermann, M. Schumm, J. Dean, and J. Schmitz. (1998) “Walknet – a biologically inspired network to control sixlegged walking,” in Neural Networks 11 (1998) 1435–1437. H. Cruse and C. Bartling. (1995) “Movement of Joint Angles in the Legs of a Walking Insect, Carausius morosus,” in J. Insect Physiol., Vol. 41, No. 9, pp. 761–771, 1995. Ö. Ekeberg, M. Blümel and A. Büschges. (2004) “Dynamic Simulation of Insect Walking,” in Arthropod Structure & Development 33 287 (14 pages), 2004. K. S. Espenschied, R. D. Quinn, H. J. Chiel, and R. D. Beer (1996). “Biologically Based Distributed Control and Local Reflexes Improve Rough Terrain Locomotion in a Hexapod Robot,” in Robotics and Autonomous Systems, vol. 18, pp. 59–64, 1996. M. Frik, M. Guddat, M. Karatas, and C. D. Losch. (1999) “A novel approach to autonomous control of walking machines,” in Proc. of the 2nd Int. Conf. on Climbing and Walking Robots (CLAWAR1999), (ed. G. S. Virk, M. Randall & D. Howard), pp. 333–342. Bury St. Edmunds, London: Professional Engineering Publishing Limited, 1999. B. Gaßmann, K.-U. Scholl, and K. Berns. (2001) “Behavior Control of LAURON III for Walking in Unstructured Terrain,” in Proc. Intl. Conference on Climbing and Walking Robots (CLAWAR’01), pp. 651– 658, Karlsruhe, Germany, September 2001. D. Hess and A. Büschges. (1997) “Sensorimotor Pathways Involved in Interjoint Reflex Action of an Insect Leg,” in Journal of Neurobiology 1997; 33(7):891-913.

[10] W. A. Lewinger and R. D. Quinn. (2010) “A Hexapod Walks Over Irregular Terrain Using a Controller Adapted from an Insect’s Nervous System,” in Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS'10): Taipei, Taiwan, October 18-22, 2010. [11] W. A. Lewinger, B. L. Rutter, M. Blümel, A. Büschges, and R. D. Quinn. (2006) “Sensory Coupled Action Switching Modules (SCASM) generate robust, adaptive stepping in legged robots,” in Proc. of the 9th Int. Conf. on Climbing and Walking Robots (CLAWAR2006), pp. 661–671, Brussels, Belgium, Sept 12–14, 2006. [12] W. A. Lewinger and R. D. Quinn. (2008) “BILL-LEGS: Low computation Emergent Gait System for Small Mobile Robots,” in Proc. of IEEE International Conference on Robotics and Automation (ICRA'08): pp. 251-256, Pasadena (CA), USA, May 11-23, 2008. [13] F. Pfeiffer, H. J. Weidemann, and J. Eltze. (1994) “The TUM Walking Machine. - In: Intelligent Automation and Soft Computing,” in Trends in Research, Development and Applications, TSI Press, vol. 2, pp. 167–174, 1994.