Fig. 3: Octopus-inspired soft robotic arm architecture and control.

a, Prototyped arm with a schematic of its physical components: peristaltic water pumps, tendon pulleys and motors and the soft arm with a close-up view at its base showing the arm rotational joint about its (straight pose) axis, the three tendons, the suction channels and the electronic and communication wires entering the arm. Four large (L), four medium (M) and two small (S) artificial suckers were integrated into the arm and grouped into four suction channels. Internally, the soft arm comprises two tendons arranged on the ventral side (green (T2) and blue (T1)) and one tendon on the dorsal side (yellow (T3)). The sensing board of each suction cup is connected to a connection board. Together, they constitute, by analogy with the natural model, the local circuitry formed by the SG and the BG. Multiple connection boards are arranged in series and connected to a master board that embeds the peripheral control (PNS) and interfaces with an external PC for debugging. b, Octopus neural circuitries, as described in the literature⁵⁷. Elements of the LNC and the peripheral neural system are highlighted and mapped onto the physical architecture in the artificial arm through dashed rectangles. c, Proposed control scheme for the robotic arm implements multiple LNC control layers running in parallel, one for each suction cup, and a single, hierarchically superior PNS layer. The LNC layer detects contact, determines the contact direction and triggers local suction (each LNC operates independently and in parallel with the other LNCs and the PNS). The PNS layer collects contact direction information from all suction cups and defines the grasping strategy, overriding the LNC’s behaviour, if needed. A compact pseudocode of the main operations performed by the PNS layer is presented in the box, along with a schematic illustrating how different grasps are discerned. PID, proportional–integral–derivative.