Practical hardware for evolvable robots

Abstract

The evolutionary robotics field offers the possibility of autonomously generating robots that are adapted to desired tasks by iteratively optimising across successive generations of robots with varying configurations until a high-performing candidate is found. The prohibitive time and cost of actually building this many robots means that most evolutionary robotics work is conducted in simulation, but to apply evolved robots to real-world problems, they must be implemented in hardware, which brings new challenges. This paper explores in detail the design of an example system for realising diverse evolved robot bodies, and specifically how this interacts with the evolutionary process. We discover that every aspect of the hardware implementation introduces constraints that change the evolutionary space, and exploring this interplay between hardware constraints and evolution is the key contribution of this paper. In simulation, any robot that can be defined by a suitable genetic representation can be implemented and evaluated, but in hardware, real-world limitations like manufacturing/assembly constraints and electrical power delivery mean that many of these robots cannot be built, or will malfunction in operation. This presents the novel challenge of how to constrain an evolutionary process within the space of evolvable phenotypes to only those regions that are practically feasible: the viable phenotype space. Methods of phenotype filtering and repair were introduced to address this, and found to degrade the diversity of the robot population and impede traversal of the exploration space. Furthermore, the degrees of freedom permitted by the hardware constraints were found to be poorly matched to the types of morphological variation that would be the most useful in the target environment. Consequently, the ability of the evolutionary process to generate robots with effective adaptations was greatly reduced. The conclusions from this are twofold. 1) Designing a hardware platform for evolving robots requires different thinking, in which all design decisions should be made with reference to their impact on the viable phenotype space. 2) It is insufficient to just evolve robots in simulation without detailed consideration of how they will be implemented in hardware, because the hardware constraints have a profound impact on the evolutionary space.

Keywords: autonomous robot fabrication; evolutionary robotics; hardware constraints; hardware design; modular robots; robot manufacturability.

FIGURE 1

The simulation domain provides a large space of possible robots that could be evolved, represented here as blue dots (top). However, only portions of this space contain robots that are practically feasible in hardware (bottom). The scatter plots representing the two landscapes are taken from experimental data presented in Buchanan et al. (2020a).

FIGURE 2

An illustration of how the hardware design paradigm changes for an evolvable robot platform. Rather than the design being defined by fixed constraints derived from the desired functionality for the robot, the hardware design comes first, and defines the constraints on the functionality available to the evolved robots.

FIGURE 3

The production process to create a robot designed by evolution, from digital to physical. The step numbers correspond to those described in detail in the main text. Firstly the skeleton must be 3D printed before the head (containing the controller and battery) is inserted. Then wheels and sensors can be attached and connected to the head via retracting cables. The complete process of producing a physical robot from its digital specification is done autonomously.

FIGURE 4

The organs available for the ARE system are the following: head, wheel, sensor, joint and castor. The controller of the robot is run in the head organ, which also supplies the other organs with power and communications. The wheel organ provides rotary locomotion. The sensor organ contains two sensors, enabling the measurement of distance from objects and the detection of infrared light. The joint organs provide powered articulation points for forming limbs, as in the pictured example of a two-jointed leg (joints in blue). The castor ball is the unique passive organ designed to reduce the friction between the robot and the floor.

FIGURE 5

This figure shows the mechanical connector used in the ARE platform. The female side (A) uses two sprung arms to engage with the indentations on the male side (B) to form the final connection (C).

FIGURE 6

Retractable cables are used to interconnect organs, shown here at the different stages of the connection process: 1) The organ is clipped onto the skeleton. 2) The retractable cable is drawn out from a cavity in the organ casing. 3) The cable is connected to the head. 4) The coiled cable self-adjusts its length to the distance between the head and the organ.

FIGURE 7

Components of the electronic hardware. The battery (A) is connected to a motherboard that is connected to the Raspberry PI (B). To the motherboard is also connected a pair of daughter boards, each of which has four TRRS sockets for organ connections (C). An example of daisy-chain topology with two joint organs can be seen across (C–E), whereby a cable from the distal joint in (E) is plugged into a second socket on the proximal joint in (D), which in turn is plugged into one of the organ sockets on the daughter board in (C).

FIGURE 8

An illustration of how load-induced voltage drops manifest at different points within the power distribution network, using the example of a daisy chain comprising two organs. The total combined current draw of all components I _TOTAL induces a drop T in the battery output voltage due to its internal resistance R _BAT . The current drawn by each organ then induces additional voltage drops A and B across the resistance of the cables, such that the supply voltage at the organ inputs is further reduced. Note that the current to Organ B must travel through both the first and second cables, so its effect is multiplied by its daisy chain position. In real operation the blocks in the diagram would dynamically expand and contract as the load varies, but when power budgeting we must allow for the worst case at peak current. If the sum of these drops at any point in the system brings the voltage below the brownout threshold, the robot will malfunction, so avoiding this is a necessary condition for reliability.

FIGURE 9

The physical equivalent of the example in Figure 8 with two joint organs in a daisy chain and other organs omitted for clarity. Current for the first joint flows from the green battery at the bottom of the head organ, up via the motherboard and daughter board and along the first organ cable. The current for the second joint follows the same path before continuing along the second organ cable, so its current must flow through both the first and second organ cables, increasing the impact of the second joint on the power budget.

FIGURE 10

The output voltage measured across the terminals of the 5-cell Ni-MH battery at different load currents. The gradient of the current-voltage line indicates the internal resistance of the battery, approximately 0.3 Ω.

FIGURE 11

The output voltage measured at the output of the boost-buck regulator circuit at different load currents and input voltages, showing that it can easily step down from higher voltages, but the extra switch current required in boost mode limits how low the input voltage can go. Sudden drops indicate thermal cutout of the regulator, showing the limits of its output capability.

FIGURE 12

Example of evolved robots. (A) This robot has 2 wheels on one side of the robot and 4 caster balls on the other side and this robot is capable of moving in a straight line by the caster balls getting stuck between the gaps of the tiles. (B) This robot has two wheels, one on each side, and four sensors. This robot has a tendency of leaning towards either side. The best behaviour can be seen if the robot leans towards the left side, however, if the robot is leaning on the right side, the robot reverts to the left side by hitting the gaps between the tiles. (C) This robot uses the bulk plastic to nudge its way around obstacles instead of avoiding them.

FIGURE 13

This diagram brings together the hardware design paradigm described in Figure 2 with the “design” space available to the evolutionary process. The hardware constraints define which regions of the complete evolvable space defined by the representation are practically feasible, and the combined result of this is a more restricted viable phenotype space.