Brain-Body-Task Co-Adaptation can Improve Autonomous Learning and Speed of Bipedal Walking

TL;DR

This study addresses autonomous bipedal locomotion by combining a tendon-driven, over-actuated leg with brain-body-environment co-adaptation. It introduces a natural motor babbling strategy that, when paired with a simple 3-layer ANN mapping 6D limb kinematics to 3 motor commands, enables locomotion after only about 2 minutes of exploration, leveraging backdrivable limb dynamics to manage interactions with the environment. Compared with naive babbling, natural babbling yields substantially higher data spread in locomotion-relevant regions and higher learning success, with 75% success under slight ground contact and 100% when trajectories are pushed below ground, accompanied by faster walking speeds. The work demonstrates that locomotion can emerge from bio-inspired co-design and co-adaptation without explicit trajectory-error control, underscoring the role of physical plant dynamics and exploration strategies in data-efficient lifelong learning for robots.

Abstract

Inspired by animals that co-adapt their brain and body to interact with the environment, we present a tendon-driven and over-actuated (i.e., n joint, n+1 actuators) bipedal robot that (i) exploits its backdrivable mechanical properties to manage body-environment interactions without explicit control, and (ii) uses a simple 3-layer neural network to learn to walk after only 2 minutes of 'natural' motor babbling (i.e., an exploration strategy that is compatible with leg and task dynamics; akin to childsplay). This brain-body collaboration first learns to produce feet cyclical movements 'in air' and, without further tuning, can produce locomotion when the biped is lowered to be in slight contact with the ground. In contrast, training with 2 minutes of 'naive' motor babbling (i.e., an exploration strategy that ignores leg task dynamics), does not produce consistent cyclical movements 'in air', and produces erratic movements and no locomotion when in slight contact with the ground. When further lowering the biped and making the desired leg trajectories reach 1cm below ground (causing the desired-vs-obtained trajectories error to be unavoidable), cyclical movements based on either natural or naive babbling presented almost equally persistent trends, and locomotion emerged with naive babbling. Therefore, we show how continual learning of walking in unforeseen circumstances can be driven by continual physical adaptation rooted in the backdrivable properties of the plant and enhanced by exploration strategies that exploit plant dynamics. Our studies also demonstrate that the bio-inspired codesign and co-adaptations of limbs and control strategies can produce locomotion without explicit control of trajectory errors.

Figures (7)

• Figure 1: A-Tendon route diagram of one leg, B- Render of the 3D model of the biped C.- Photograph of the tendon-driven bipedal robot. To reduce rotational inertia, motors M1 and M2 (Maxon DCX16S GB KL 24V, 21:1 reduction ratio gearhead) are placed distally to the joints.
• Figure 2: Representation of the ANN used as a map from six limb kinematics (input nodes, left column) to 3 motor activations (output nodes, right column). In this figure we show real data used to train the ANN (particularly naïve babbling data) As a reminder, motor activations in babbling are random (Specific details on naïve babbling are given in sections:\ref{['methods:naive_natural']}). The ANN has three fully connected layers: input, hidden and output layers with respectively 6, 15 and 3 nodes. Note that motors are persistently simultaneously activated (i.e., coactivation, see motors M2 and M3 in leftmost panel), this decreases the spread in training data.
• Figure 3: Representation of the ANN used as a map from six limb kinematics (input nodes, left column) to 3 motor activations (output nodes, right column). In this figure we show real data used to train the ANN (particularly natural babbling data) As a reminder, motor activations in babbling are random (Specific details on natural babbling are given in section:\ref{['methods:natural']}). The ANN has three fully connected layers: input, hidden and output layers with respectively 6, 15 and 3 nodes. This figure shows how oscillatory movements are produced (see "Normalized angular positions" panel) driven by the by oscillatory babbling activations (rightmost panel) with significant difference activation level between antagonist motors.
• Figure 4: Desired joint and foot trajectories. In A it is shown a hip and knee joint evolution profile that produces limb movements away from the limits of its in-air configuration space as shown in B (Both panels in B represent the same, right one is a zoom out version). The resultant foot trajectory, shown in C is such that permits its front and back swings to have different height (necessary point to produce locomotion). Note that the desired foot trajectory is always kept at a constant distance from the hip; thus if the hip position is changed the desired foot trajectory will also change.
• Figure 5: Two minutes of babbling data and desired trajectories for one trial. A: Joint Space with joint motion limits marked with a doted square, B: Endpoint Space. The spread values are 0.14 and 0.18 for naïve babbling data (left and right legs respectively) and 0.60 and 0.53 for natural babbling data (left and right legs respectively)