Chapter Four · failure evidence
What Actuation & Robotic Control got wrong, from 26 dissertations
The records document experimental failures, baseline comparisons, and rejected designs across robotic actuation, physical mechanics, and control architectures. Investigators frequently encountered difficulties with soft actuator fabrication and efficiency, mechanical hysteresis in continuum transmissions, dynamic contact instabilities, and unconstrained policy actions. These records come from PhD theses at 8 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Soft and fluidic actuators suffer from low efficiency, material breakdown, and structural asymmetry
Soft pneumatic and hydrogel actuators displayed poor energetic efficiency compared to rigid baselines, while hyperelastic constitutive models broke down numerically under high actuation pressures. Overlapping fabric seams caused asymmetric stiffness and unintended bending, and sharing a single pneumatic chamber caused shape-changing pressures to corrupt tactile sensing.
Tried and failed
finite element analysis with hyperelastic material models applied to pneumatic soft continuum robot grasping. Outcome: did not converge. Reason: numerical instability and breakdown of hyperelastic constitutive models under high actuation pressures
Modeling, Planning, and Learning for Soft Robots in Human-Centric, Contact-Rich Environments · Harvard
Tried and failed
single-chamber simultaneous pneumatic actuation and sensing applied to soft robotic end-effectors. Reason: actuation pressure changes for shape-changing interfere with using the same cavity for tactile pressure sensing
A soft robotic approach for robot-assisted palpation training · Imperial
Tried and failed
fabric tube reinforced soft pneumatic actuators applied to burrowing soft robot propulsion. Reason: overlapping fabric seams created asymmetric stiffness causing unintended non-uniform bending during actuation
Micromechanics of passive and active inclusions in granular media · Georgia Tech
Lost to a baseline
Soft motor efficiency of 0.24% to 0.35% was lower than simulated rigid equivalent motor (~70%) and commercial hard motors
The Design and Application of Soft Electromagnetic Actuators · Georgia Tech
Lost to a baseline
Porous sweating PAAm hydrogel actuators demonstrated lower actuation efficiency while actively sweating/thermoregulating.
THE SENSITIVE SOFT ROBOT FROM MULTIMODAL SKIN TO SELF-HEALING ADAPTIBILITY · Cornell
Mechanical hysteresis, friction, and flexibility degrade cable and continuum actuation
Linear displacement models and kinematic pose tracking failed due to direction-dependent friction and mechanical hysteresis between forward and return cycles. Cable flexibility, lack of direct joint angle observability, and bulky support equipment also prompted the rejection of soft pneumatic and cable-driven mechanisms.
Tried and failed
linear displacement model for actuator positioning applied to robotic actuator displacement prediction. Outcome: worse than baseline. Reason: nonlinear friction and direction-dependent actuation differences caused severe prediction errors
Image Guided High Precision Robotic Positioning in MRI for Medical Applications · Georgia Tech
Tried and failed
kinematic pose tracking from tendon displacement applied to continuum robot position estimation. Reason: mechanical hysteresis caused path misalignment between forward and return actuation cycles
Forward-Viewing Ultrasound Guidance of a Robotically-Steered Guidewire for Peripheral Interventions · Georgia Tech
Considered and rejected
Considered and rejected: Rejected cable-driven piezoelectric ultrasonic motor mechanisms due to cable flexibility and hysteresis.
Image Guided High Precision Robotic Positioning in MRI for Medical Applications · Georgia Tech
Considered and rejected
Considered and rejected: Rejected pneumatic and cable-driven soft actuation due to cumbersome pumps/tanks, lack of direct joint angle observability, and inability to independently control multi-DOF joint trajectories without external object contact.
Towards Naturalistic Exoskeleton Glove Control for Rehabilitation and Assistance · Virginia Tech
Contact dynamics and unmodeled disturbances induce control instability
Exoskeleton impedance controllers and open-loop resonant microrobots experienced severe oscillations and heading drift due to non-rigid interface resonance and minute surface defects. Bipedal operational space controllers and actuator-level command cancellation similarly failed because of impact timing sensitivity and internal instability under integral feedback.
Tried and failed
servomotor-based passive impedance tremor suppression applied to wearable wrist exoskeleton. Outcome: unstable. Reason: non-rigid interface contact dynamics caused phase shifts and resonance, amplifying tremor oscillations instead of suppressing them
Design and Control of an Ergonomic Wearable Full-Wrist Exoskeleton for Pathological Tremor Alleviation · Virginia Tech
Tried and failed
open-loop resonant vibration actuation for locomotion applied to microrobot trajectory steering. Outcome: unstable. Reason: minute surface defects and unlevelness caused stochastic heading drift and trajectory discontinuities
Sensing, actuation, and intelligence integration in micro-robotics · Georgia Tech
Tried and failed
operational space control without impact projection applied to bipedal running robot control. Outcome: unstable. Reason: severe underactuation and velocity error spikes from impact timing sensitivity
Considered and rejected
Considered and rejected: Rejected cancelling expected motor commands at the actuator level rather than sensory signals because it exhibits internal instability under integral feedback and fails to prevent sensor saturation.
Bio-inspired robotics: efference copies and adaptive feedforward control · Imperial
Fixed or minimal aerial actuation setups lack fault tolerance and gust rejection authority
Standard alternating-direction multirotor configurations failed to maintain yaw control authority following motor or processor failures, prompting the adoption of redundant hexacopter frames. Additionally, single-surface pitch control without coordinated dihedral actuation produced severe vertical accelerations during high-frequency vertical gusts.
Tried and failed
standard alternating-direction multirotor propeller configuration applied to fault-tolerant hexacopter control. Outcome: unstable. Reason: loss of yaw control authority following single motor or processor failure
Optimization-based design of fault-tolerant avionics · Georgia Tech
Tried and failed
single-surface pitch control without coordinated dihedral actuation applied to high-frequency vertical gust rejection. Outcome: worse than baseline. Reason: elevator-only control yielded large vertical accelerations compared to combining pitch with active dihedral morphing
Multibody Dynamics of Aeromechanical Systems · Virginia Tech
Considered and rejected
Considered and rejected: Rejected quadcopter configuration in favor of hexacopter (hexa+) to ensure motor-failure redundancy, reduced vibration, and proper arm alignment with crop rows.
Design of an Innovative Spraying System for High Precision Aerial Dispersion in Vineyards · IRIS - POLITO - prod
Unregularized or unconstrained reinforcement learning yields erratic robot actions
Residual policies without action regularization produced excessively large actions that damaged trajectory quality and prevented sim-to-real transfer. Similarly, directly activating internal latent units yielded erratic dashes instead of structured discrete motor actions.
Tried and failed
residual policy learning without action regularization applied to robot motor skill composition. Outcome: did not generalise. Reason: unregularized residual actions grew excessively large, degrading trajectory quality and preventing sim-to-real transfer
Learning interactive robot behaviors using deep reinforcement learning · Georgia Tech
Tried and failed
latent unit activation for behavioral control applied to biomechanical reinforcement learning policy. Outcome: did not generalise. Reason: activating internal representation units yielded erratic dashes rather than structured discrete motor actions
Modeling the neural control of movement with a virtual rodent · Harvard
Left open by the authors
Problems the authors named and did not get to.
Left open
Deploy and validate the developed sampling-based Model Predictive Control (MPC) frameworks on real-world robotic hardware. Blocker: Requires access to physical robotic hardware systems and an experimental robotics lab setup.
Sampling-based Dynamic Optimization: Theory, Analysis and Applications · Georgia Tech
Left open
Test motor averaging versus performance-optimization on intermediate reaching movements under forced response times below 222ms using a robotic manipulandum setup. Blocker: Requires human subjects and experimental hardware (robotic manipulandum/vBOT reaching apparatus)
Understanding How Environmental Dynamics and Uncertainty Affect Adaptive Changes in Motor Planning · Harvard
Left open
Evaluate parallel trajectory planning performance on offloaded mobile robotics SLAM workloads and multi-robot physical testbeds. Blocker: Requires access to multiple physical manipulator robots and mobile robotic hardware setups
MATRIX: Manipulator Avoidance Trajectories via Reactive, Integrated eXtreme-scale computing · Queens University Institutional Repository
Left open
Train a neural network conditioned on previous robot states and actions to model real-world actuation noise for sim-to-real transfer. Blocker: Requires real robot hardware trajectory data to model real-world actuation dynamics.
Sim2Robot: Training Robots for the Real-World with Imperfect Simulators · Georgia Tech
Left open
Implement a rate-dependent or multi-axis vector Preisach hysteresis model parameterized across actuation speeds for soft robotic actuators. Blocker: Requires a physical soft actuator apparatus to collect multi-speed and multi-axis characterization data.
Mechanics of Soft-Rigid Interactions: From Wearable Robots to Stingray Skeletons · Harvard
Left open
Integrate selectively stiffening microchannels or dielectric elastomer actuators into elastomer domes to reprogram deflection directions for omnidirectional actuation. Blocker: Requires a physical soft robotics fabrication laboratory, specialized elastomer materials, and testing apparatus
Manufacturing and Design of Echinoderm-Inspired Underwater Soft Robots · Harvard
Left open
Integrate soft logic valves directly inside soft robot appendages to replace discrete solenoid valves for decentralized control. Blocker: Requires a physical soft robotics manufacturing and testing lab to fabricate and validate the soft logic valves and appendages.
Manufacturing and Design of Echinoderm-Inspired Underwater Soft Robots · Harvard
Left open
Incorporate embedded soft sensors directly into DEA-driven pump-valve structures for closed-loop fluidic feedback. Blocker: Requires soft robotics fabrication lab, dielectric elastomer actuators, elastomeric channels, and physical testing hardware
Left open
Develop a soft robotic hand with an integrated 1-DOF mechanism to dynamically adjust digit arrangements on the fly. Blocker: Requires mechanical engineering hardware fabrication and a robotics lab to construct and test the physical soft gripper mechanism.
Design Principles for Improving Precision and Dexterity of Soft Robotic Hands · Harvard
Left open
Investigate dielectric elastomer actuators or electroactive polymers to replace pneumatic joint stiffening in soft flexure robotic joints. Blocker: Requires a physical materials/robotics lab and fabrication hardware to test dielectric elastomer actuators.
Characterization of Soft Structures for Synergy-Based Performance Optimization in Robotic Grasping and Other Manipulation Tasks · Georgia Tech
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.