Biological and engineered systems both depend on precise internal sensing to maintain balance, adapt to damage, and continue moving when components fail. Research in physiology and robotics converges on the conclusion that no single sensor suffices for robust, fault-tolerant legged locomotion. Instead, a layered and redundant suite of proprioceptive modalities is most effective.
Biological foundations
Classic work by Peter Proske University of Adelaide and Simon Gandevia University of New South Wales emphasizes the complementary roles of muscle spindles and Golgi tendon organs in encoding muscle length and force. These sensors enable rapid spinal and supraspinal reflexes that correct perturbations before conscious control intervenes. Joint receptors and cutaneous mechanoreceptors in the foot add geometric and contact information that informs posture and gait. This diversity creates natural redundancy so that loss or corruption of one signal can be compensated by others.
Robotic implementations
Legged robotics research led by Sangbae Kim Massachusetts Institute of Technology and field practices from Marc Raibert Boston Dynamics show parallel design choices. Robots rely on high-resolution joint encoders for kinematics, torque or force sensors at joints and limbs for loading information, and inertial measurement units for body orientation and rate feedback. Foot contact sensors or pressure-distributed sensors provide discrete ground truth about stance. Modern controllers fuse these streams to detect sensor faults, reweight estimators, and switch control modes when necessary.
Fault-tolerance principles and consequences
The most fault-tolerant architectures combine redundancy, heterogeneity, and distributed sensing. Redundancy means multiple sensors measure overlapping quantities so a failing sensor can be isolated. Heterogeneity pairs rapid local reflex-like signals from force or stretch sensors with slower, global IMU and model-based estimators to preserve stability during complex failures. Distributed sensing places basic reflex loops at each joint or limb to maintain local stability even when central computation is impaired. Culturally and environmentally this matters for deployment in remote or hazardous regions where maintenance is difficult and human rescue may be delayed.
Implementing these principles reduces catastrophic falls, extends operational life in harsh terrains, and lowers reliance on perfect communications. For practitioners the takeaway is clear: prioritize a mixed sensor suite that mirrors biological proprioception, design fusion and fault-detection algorithms, and locate simple reflexive controllers close to limbs to preserve locomotion under partial failure.