Engineering simulators for robots, scanners, agents & learning machines

Tune the position and velocity loops of a physics-based DC servo. Includes live scopes, step-response metrics, relay auto-tuning, Bode stability margins and closed-loop pole maps, with encoder quantisation, friction and saturation modelled.

Drag a 2- or 3-link manipulator, solve inverse kinematics analytically or with damped least squares, see the configuration-space obstacles and plan collision-free, smoothly time-scaled motions around obstacles.

Scan analytic phantoms, lower the dose to add Poisson noise, introduce metal and motion artefacts, then reconstruct with filtered back-projection or iterative SART with total-variation regularisation. Error maps and PSNR/SSIM quantify the result.

Set spin-echo, inversion-recovery and gradient-echo contrasts on a synthetic brain. Paint directly on k-space, simulate motion and spike artefacts, and recover undersampled scans with compressed sensing and partial-Fourier reconstruction.

A powered wheelchair with LIDAR plans cost-aware routes through an apartment and follows them with a dynamic-window planner. It takes voice-style commands, blends in the joystick of a user with tremor, dodges people and builds a map as it drives.

Synthesise 8-channel forearm EMG, filter it, extract time-domain features, train LDA, k-NN or neural-network classifiers and drive an animated robotic hand in real time, with confusion matrices and feature-space projections.

Build and train a multilayer perceptron in your browser. Watch each neuron's activation map, the decision boundary and the per-layer gradient flow, and run experiments on vanishing gradients, overfitting and feature engineering.

A population of neural-network drivers learns to race on procedurally generated circuits through a genetic algorithm. They have ray sensors, grip-limited handling and a live brain view. Test whether they generalise to new tracks, then race the champion yourself.

Design LLM-agent pipelines as graphs: planners, parallel tool calls, reflection loops, routers and human approval. A discrete-event simulator then measures latency, token cost and reliability under retries, timeouts, fallbacks, rate limits and caching, and critical-path analysis shows where to optimise.

Run random-effects meta-analyses (DerSimonian–Laird, REML, Knapp–Hartung) with forest and contour-enhanced funnel plots, Egger's test and trim-and-fill, plan studies with exact power analysis, and simulate how publication bias and p-hacking distort a literature.

Drive an electric car through city, highway and sporty cycles on a 96-cell pack of equivalent-circuit Li-ion cells. Compare state-of-charge estimation by coulomb counting, voltage lookup and extended Kalman filters (including a current-sensor-bias state), then charge with CC-CV and watch passive balancing recover capacity stranded by the weakest cell.

A wearable accelerometer streams daily life and falls of an older adult. Four detectors (impact threshold, posture checks, a free-fall state machine and a learned model) drive an "Are you OK?" alert and caregiver escalation. A benchmark then shows why algorithms tuned on simulated lab falls miss real ones, trading sensitivity against false alarms per day.