Issue 41 Unmanned Systems Technology December/January 2022 PteroDynamics X-P4 l Sense & avoid l 4Front Robotics Cricket l Autonomous transport l NWFC-1500 fuel cell l DroneX report l OceanScout I Composites I DSEI 2021 report

The Cricket uses revolute joints with high degrees of articulation, with the ankles for example able to pivot the tracks freely through n x 360 º , so each track can be at a different height and angle 52 Digest | 4Front Robotics Cricket of motion. Each limb has two shoulder joints, with the first pivoting on an axis parallel to the body’s fore and aft axis and the freedom to move through 180 º (±90 º ), with the second pivoting through 350 º on an axis at 90 º to the first. The knee and ankle joints are free to rotate in the same sense as the second shoulder joint but in slightly offset planes and without angular limits (n x 360 º ). The robot uses a full-body motion algorithm to determine the best pose to adopt to move through any given space based on the direction of motion, the tasks it has to perform and the configurations of the environment. This is supported by a custom sensing algorithm that evaluates the complexities of the terrain and determines how to maintain stability while moving through it, using a model predictive control approach combined with deep reinforcement learning algorithms. Such decisions are based on geometrical features of the terrain, such as slope and roughness, along with other perceived physical terrain characteristics such as hardness, friction and compliance. From manual to AI control The Cricket can be controlled manually, semi-autonomously and autonomously, with the decision on which type of control use depending greatly on the specifics of the task in hand. Simple locomotion tasks such as skid steering with the tracks can easily be executed manually, Dr Ramirez-Serrano explains. “However, tasks requiring continuous motion of all the joints, such as climbing a ladder, benefit from autonomous control, although they can be executed under semi-autonomous control where the user sends only speed and direction commands to the robot, which then computes how to carry them out,” he says. He adds that the onboard control system makes extensive use of AI to generate motion commands when the Cricket needs to move in unusual ways. Two Nvidia Jetson Xavier units provide the computing power for the control, navigation, path planning and sensing tasks. There is also the option of control via a tether, which allows power to be sent to the vehicle as well, extending its endurance from around 3 hours on the battery to technically unlimited. This can be useful in missions involving a lot of climbing, which is particularly power-hungry. During its development, the control system has undergone many iterations, Dr Ramirez-Serrano says, with testing of numerous formal (or classical) and AI (machine learning) algorithms. His team has also used many of the modelling approaches common in robotics, but is also working to move beyond them. “Owing to the complexities and associated uncertainties in the mathematical models – there is no perfect model for any system – we are developing model-free control and navigational approaches in which the robot doesn’t depend on the model for error-free operation,” he explains. Also in prospect are two new versions of the Cricket, the first fitted with more compliant variable joints to improve the robot’s mobility in complex terrain, the second being a double-sized version with hydraulic actuators to allow it to take on very power-hungry applications. December/January 2022 | Unmanned Systems Technology Overall length: 630 mm Body length: 410 mm Body width: 360 mm Track depth: 60 mm Track length: 250 mm Ground speed: 0.5 m/s Sensor and effector options: cameras, manipulator arms, fire extinguishers and others Some key suppliers DC motors: Faulhaber Drive Systems Batteries: Grin Technologies Actuators: Hebi Robotics Vision sensors: Intel NUC computers: Intel Stepper motors: Maxon Precision AI computers: Nvidia Specifications (electric version)