Bio-Inspired Computing for Robotics

Translate a robotics capability into a measurable bio-inspired computing benchmark.

Define performance, robustness, adaptation-speed, compute, and energy metrics for a robotics experiment.

Create a reusable Python and ROS2 experiment scaffold with configurable controllers and environments.

Record configurations, random seeds, trajectories, and timing data for reproducible experiments.

Implement a conventional controller as a baseline for a bio-inspired method.

Quantify uncertainty in repeated robot experiments using confidence intervals and effect sizes.

Represent a robot controller as a genotype-to-phenotype mapping suitable for black-box optimization.

Implement an ask-tell evolution strategy for optimizing controller parameters.

Apply CMA-ES to a noisy robotics objective.

Design a behavior descriptor that separates meaningfully different robot behaviors.

Implement novelty search using a nearest-neighbor novelty metric.

Implement MAP-Elites for a two-dimensional behavior archive.

Apply an emitter-based quality-diversity method to improve archive coverage and quality.

Select robust controllers from a quality-diversity archive using perturbed evaluations.

Encode a feedforward neural controller for direct evolutionary optimization.

Implement topology and weight mutations for a NEAT-style controller.

Apply speciation to preserve structural innovation during neuroevolution.

Evolve a recurrent neural controller for a partially observable robotics task.

Construct a CPPN-based indirect encoding for a structured robot controller.

Formulate a multi-objective neuroevolution problem balancing task performance and resource use.

Measure the generalization of evolved policies across unseen environments.

Diagnose deceptive objectives by comparing fitness-driven and diversity-driven neuroevolution.

Simulate a nonlinear oscillator and identify its stable limit cycle.

Implement a Hopf oscillator as a rhythmic motor command generator.

Couple multiple oscillators with specified phase relationships.

Map oscillator states to joint trajectories for a simulated robot.

Tune oscillator parameters to switch between distinct locomotion patterns.

Modulate a central pattern generator using proprioceptive feedback.

Optimize central pattern generator parameters with an evolution strategy.

Evaluate gait recovery after external perturbations or simulated actuator damage.

Simulate a leaky integrate-and-fire neuron under time-varying input.

Encode continuous sensor values using rate and temporal spike codes.

Decode spike trains into continuous motor commands.

Implement spike-timing-dependent plasticity for a small spiking network.

Stabilize spiking activity with a homeostatic adaptation rule.

Train a spiking network with surrogate gradients on a sensor-classification task.

Construct a spiking policy for a closed-loop robotics task.

Compare spiking and non-spiking models by accuracy, latency, operation count, and estimated energy.

Transform an event-camera stream into time surfaces and voxel-grid representations.

Filter background activity and refractory noise from an event stream.

Estimate image-plane motion from asynchronous events.

Generate controlled event-camera data in a robotics simulator.

Train a spiking model to recognize motion patterns from simulated events.

Connect event-based perception to a reactive robot controller.

Measure end-to-end latency and bandwidth under frame-based and event-based sensing.

Partition an event-processing pipeline between software and FPGA-oriented streaming blocks.

Construct an echo-state network with stable reservoir dynamics.

Train a reservoir readout for robot state prediction.

Implement a continuous-time recurrent neural controller.

Implement a closed-form continuous-time neural policy for irregularly sampled sensor data.

Interpret the sparse wiring of a neural circuit policy in terms of sensor-to-action pathways.

Compare recurrent, liquid, and spiking policies under sensor noise and timing jitter.

Combine a learned dynamical policy with a central pattern generator through bounded modulation.

Implement a decentralized swarm controller using only local observations.

Produce cohesive flocking by balancing separation, alignment, and attraction rules.

Implement distributed consensus over a time-varying communication graph.

Allocate tasks among robots using response thresholds or local auctions.

Implement stigmergic coordination through a simulated environmental signal.

Prevent inter-robot collisions without introducing centralized planning.

Evolve or train a shared local policy for an entire homogeneous swarm.

Measure swarm scalability and resilience under communication loss and robot failures.

Identify how compliance and passive dynamics can reduce a controller's computational burden.

Model a compliant actuator or elastic joint in simulation.

Measure the contribution of passive body dynamics by comparing locked and compliant configurations.

Define morphology descriptors for a quality-diversity archive.

Co-optimize selected body parameters and controller parameters for a locomotion task.

Evaluate a body–controller design across terrain and payload variations.

Use archive diversity to select a design that can compensate for simulated structural damage.

Represent a robot's beliefs and observations with a compact generative model.

Compute prediction errors and belief updates in a discrete active-inference model.

Select actions by minimizing expected free energy in a small decision task.

Implement predictive-error control for a continuous robot state variable.

Add a homeostatic variable that changes a robot's action preferences.

Implement curiosity-driven exploration using prediction error or information gain.

Compare active-inference and reward-driven agents on adaptation and exploration metrics.

Inject realistic sensor noise, latency, packet loss, and actuator uncertainty into simulation.

Apply domain randomization to a bio-inspired controller during optimization or training.

Use system-identification data to narrow a simulation-to-reality gap.

Impose safety and actuator constraints on the output of an adaptive controller.

Package a trained bio-inspired controller as a ROS2 component with deterministic interfaces.

Profile controller latency, memory use, and energy proxies on an embedded target.

Map a central pattern generator or spiking kernel to fixed-point arithmetic.

Design an FPGA data path for a latency-critical neuromorphic computation.

Formulate a testable capstone hypothesis for a mobile robot, UAV, manipulator, embedded platform, or swarm.

Select complementary bio-inspired components using task requirements and deployment constraints.

Specify interfaces among perception, adaptation, control, and safety components.

Integrate one perception component with one bio-inspired controller in ROS2 simulation.

Add online adaptation or quality-diversity-based controller selection to the integrated system.

Perform an ablation study that isolates the contribution of each bio-inspired component.

Stress-test the system under environmental shifts, component faults, and compute constraints.

Write a reproducible technical report that relates experimental evidence to the capstone hypothesis.