El “DIF ESTATAL” a través de su representante legal declara que:

As mentioned in section 2.5, the controller communicates with the simulation via TCP/IP whereas the joint drives presented in section 2.3 use a differential serial in- terface for communication. In the real robot bus masters are used to transmit messages received via USB from the computer to the serial interface and vice versa (Schneider et al. 2012). This is depicted in fig.2.15. To unify the interface between the walking controller and the simulation or real robot hardware, an additional layer, the BioFlex

Figure 2.14.: Visualization of the simulated robot walking on flat terrain. The housings of the body are represented by trimeshes, the leg actuators by cylinders, and the connecting leg segments by capsules. By simplifying the representation of the detailed leg housing shown in section 2.4.1 the computational load can be reduced since collision detection for capsules is computationally less complex than for trimeshes. For the same reason shadows are only illustrated for those parts of the robot that are represented by simple geometric primitives. Since the body housings are represented by trimeshes the corresponding shadows are not visualized.

server, was introduced that manages the routing of messages between the TCP/IP in- terface and the USB ports. In the protocol defined for the BioFlex bus, the first two bytes of every message are reserved for the destination and the source identifier num- ber.6 _{With distinct IDs for the TCP/IP-connected controllers and the USB-connected} bus masters and clients, the server is able to route messages in both directions. Keeping track of the requests sent to the clients and the corresponding replies, the server also detects packet loss and automatically re-requests a reply.

As the simulator does not always run in real-time, the walking controller is synchro- nized to the virtual time of the simulation. Thus, the controller does not use any timing on its own but relies completely on the external timing provided by the simulation. When connected to the real robot, this task must be performed by the server to keep

hind segment walking controller like WALKNET, fixed gait or CPG L1 L2 L3 R2 R3 R1 other processes like, e.g., data loggers, external visualization BioFlex server USB robot simulation real robot communication middleware timer TCP/IP message routing middle segment front segment BioFlex bus

Figure 2.15.: Depiction of the communication framework for controlling the real robot or the virtual robot in the simulation. The robot controllers connect via TCP/IP either to the simulation or—if the real robot is to be used—the BioFlex server. In the latter case, the messages sent from the controllers are forwarded either to the timer module or to the respective BioFlex master viaUSB. In the last step, the message is sent to the client via the BioFlex bus. If an answer is required the client can send it to the controller using the reverse process.

the interface identical. Since real-time execution is inevitable for the operation of a real robot, the server uses the computer’s hardware clock for the timing.

Using this communication schema, new walking controllers can be tested first with the simulation and then, without any modifications, on the real robot.

Summary

To test the applicability of bioinspired leg coordination for technical systems, a robotic platform is required that resembles the biological model as close as possible. However, due to the limitations of current technical fabrication, not all features can be transferred. Therefore, the features considered to be most significant must be selected. In the design of the six-legged robotHECTORespecially the morphology of the model insect Carausius

morosus was adopted since this was considered to be most relevant for replication of

insect-like walking behavior. Although an exact scaling of the stick insect measures was not accomplished—partially due to technical/mechanical requirements, partially to simplify the setup—the robot is considered to be suitable to test bioinspired control approaches.

Beside the insect-like morphology, also the compliance of the biological structures was transferred to the robot. Whereas the compliance in the insect joints is mostly due to the elasticity of the muscle fibers, in the robot the compliance was implemented by introduction of custom elastomer couplings in the joint drives. The resulting inherent compliance can be utilized for passive terrain adaption. Moreover, based on the torsion of the coupling the load in the drives can be estimated, which can be used to detect ground contact or collisions during swing phase. Since a permanent, durable connection between the elastomer of the coupling and the metal support structures could not be realized, currently an unbonded elastomer inlay is used in the joint drives. Although the inlay was manufactured with oversize, the development of backlash over a longer period of use cannot be precluded. During operation of the robot, however, no backlash was detectable.

Among the biological features that were not replicated in the robot are the adhesive pads and claws at the insects’ tarsi that are used to hold on to the substrate. These features were explicitly omitted since they cannot be used to cling to all substrates (e.g, loose sand or gravel). Aiming at substrate-independent functionality of the robot, maintenance of static stability is a prime requirement for walking locomotion to avoid tilting. Technical equivalents of the adhesive pads and claws have therefore been consid- ered redundant for fundamental robot operation. However, to increase the ability of the robot to climb steep slopes, a subsequent expansion of the legs by adhesive or gripping mechanisms might be required.

To simplify the development and evaluation of walking controllers, a simulation frame- work was developed that allows for preliminary tests on a virtual robot. For example, in chapter 4, the robot simulation is used to evaluate the walking controller WALKNET.

Also for the development of the novel, adaptable controller (see chapter5) that considers the differences between model and robot, in a first step, the simulation framework was

employed (see chapter 6). In chapter 7, finally, this newly developed controller is used for the control of the actual robot in different walking scenarios.

3. Evaluation of Static Stability of Walking