Robotics, Mechatronics & Automation

Highly specialised requests are first checked for feasibility from this information. Please do not send confidential data through the public form; describe it first and agree secure transfer afterwards.

1. 01Enquiry and requirement assessment
2. 02Technical scoping and proposal
3. 03Execution
4. 04Quality review and delivery, with an agreed round of revisions

Read the full project workflow.

In depth: how the work is done

1. Kinematic and dynamic modelling

   Kinematics relates joint variables to end-effector pose and velocity, using conventions such as Denavit-Hartenberg parameters or product-of-exponentials formulations, and addresses singularities and workspace limits. Dynamics adds inertia, Coriolis, gravity and friction terms through Lagrangian or Newton-Euler methods. We help derive models, check them numerically against simple limiting cases and identify parameters, such as link inertias and joint friction, from experiments, noting which parameters are poorly identifiable from the available motions.

2. Control design and simulation studies

   Control strategies range from PID and computed-torque control to impedance, adaptive, robust and model-predictive schemes. A fair study tunes every compared controller with comparable effort, specifies the same reference trajectories and disturbances and measures the same performance indices, such as tracking error, settling time, control effort and robustness to parameter change. Simulations include sensor noise, actuator saturation and sampling effects, because ideal assumptions often hide the problems that appear in hardware.

3. Sensing, estimation and data analysis

   Encoders, inertial units, force sensors, cameras and lidar all have noise, bias and latency. State estimation, for example with Kalman-type filters, fuses them, and its performance depends on correct noise models. We analyse recorded data to characterise sensor noise, estimate delays and evaluate estimator accuracy against a reference measurement, and we check that filter consistency statistics agree with the assumed noise levels.

4. Experimental evaluation methodology

   Hardware experiments should state the number of trials, the variation in initial conditions and payloads, and how failures were counted. Results are reported with means, spread and confidence intervals, and success rates are given with their uncertainty. Where learning-based methods are used, training and test conditions are separated, and results are shown across random seeds. A clear description of the platform, firmware and calibration allows other groups to reproduce comparable results.

5. Safety, ethics and reporting

   Robots working near people raise safety and ethics questions that research reports should acknowledge. We help describe the safety measures used in experiments, such as emergency stops and force limits, and avoid claims of safe human interaction that were not tested. Reports separate laboratory demonstration from readiness for deployment, which would require standards-based risk assessment by qualified professionals.

6. Reproducibility in robotics research

   Robotics results are notoriously hard to reproduce because platforms differ. We encourage reporting hardware models, firmware versions, control rates, calibration procedures and software dependencies, and sharing configuration files and logged trajectories where possible. Simulation environments should be versioned, and random seeds recorded. When the same method is compared with published work, the original conditions should be matched as closely as possible, and any deviation described. This gives readers a fair basis for judging whether an improvement is real and likely to transfer.

Common pitfalls this service helps avoid

• Comparing against an untuned baseline. Gains over a poorly tuned controller do not show that the new method is better.
• Evaluating on one trajectory. Performance may not carry over to other motions, payloads or speeds.
• Ignoring latency and saturation. Ideal simulations can hide instability that appears on hardware.
• Reporting success without trial counts. A success rate means little without the number of trials and conditions.

What good reporting looks like

A robotics report describes the platform and its parameters, the models and their derivation or identification, the controllers with tuning procedure, the simulation and experimental protocols, the trial numbers, the performance indices with statistical summaries, failure cases and the limitations. It does not certify a system as safe for use around people, which requires formal risk assessment under applicable standards.

An illustrative scenario

Hypothetical example written for explanation. It is not a client case, a result or a testimonial.

Imagine a researcher proposing a new adaptive controller for a two-link arm. The review finds that the baseline PID gains were set once and never tuned for the heavier payload. After retuning the baseline fairly and repeating the trials across three payloads and ten trajectories, the advantage of the adaptive controller shrinks but remains clear at the heaviest payload. The report gives tracking error with confidence intervals and notes that joint friction was identified only at low speeds.

Questions to settle before you begin

• Are my baselines tuned as carefully as my proposed method?
• Which payloads, speeds and trajectories must the claim cover?
• How noisy and delayed are my sensors?
• How many trials do I need to support my success-rate claim?
• What safety measures were used in the experiments?

Key terms

Kinematics

The geometry of motion without regard to forces.

Computed-torque control

A method that cancels modelled dynamics so the remaining system is linear.

Kalman filter

A recursive estimator that fuses a model with noisy measurements.

Singularity

A configuration where a manipulator loses a degree of freedom.

Backlash

Play in a transmission that causes delay and error on reversal.

Working together on Mechanical, Manufacturing & Industrial Engineering projects

Mechanical, manufacturing and industrial engineering engagements begin with the process or machine, the performance measures of interest and the data or models already available. We ask for process descriptions, parameter ranges, measurements and constraints, and for the decision the study should inform. Designed experiments, statistical analysis, simulation or optimisation are then selected to suit. Reports include assumptions, uncertainty and practical recommendations, and they distinguish findings that are supported by data from those that are engineering judgement.

Few projects need a single service in isolation. These are the services most often combined with this one.

• Signal Processing & Sensor Analytics

  Processing and interpretation of sensor and instrumentation signals: filtering, spectral and time-frequency analysis and feature extraction for condition assessment and experiments.

• Electronics, Embedded Systems & Instrumentation

  Analysis support for electronic, embedded and instrumentation research, including measurement-chain modelling, sensor characterisation and data-acquisition performance.

• Computational Engineering

  Application of numerical and computational methods to engineering problems where closed-form solutions are not available, with emphasis on model formulation, verification and validation.

• AI & Machine Learning for Engineering

  Machine-learning methods for engineering problems with physical data, designed around data quality, leakage-free validation and a clear account of when the model should not be trusted.