Robotics · Intelligent systems · Creative engineering

Abhinav MayuraGraduate portfolio.

Hi! This is a little collection of projects I've worked on, things I've learned and ideas I'm trying out along the way.

I’m a Mechatronics and Robotics MEng graduate from the University of Sheffield. I’m outgoing, curious and always keen to learn something new. I’ve worked on robotics, AI and hands-on engineering projects, including a hardware development internship at Amazon, using skills such as Python, simulation and CAD. I enjoy working with others, tackling new challenges and finding ways to make a useful contribution.

Explore projects ↗
SYSTEM // NAV_SIM_01● SIMULATION ACTIVE
CLICK TO SET CUSTOM WAYPOINT
LIDAR / 360°MAPPING WORLD
01 / ROBOTICS02 / MACHINE LEARNING03 / SIMULATION04 / CREATIVE SYSTEMS
// Selected work

Projects & experiments.

Completed research

Predicting robotic grasp slip

Built a simulation-to-real robot pipeline and trained an LSTM to predict when an object would slip during pick-and-place tasks.

Isaac SimLSTMSim2RealManipulation
Industry experience

Industrial robotics development

A 10-month hardware internship covering early design, technical documents, site deployment, testing and work across several teams.

CADDeploymentProject delivery
Completed control project

Rapid control prototyping

Modelled and controlled a dual-rotor helicopter using state-space control, LQR, Kalman filtering and tests on the physical system.

MATLAB / SimulinkLQRKalman filter
Completed ML project

Predicting biodegradability

Built a model to predict whether chemicals are biodegradable, then compared KNN and SVM after cleaning the data and applying PCA.

Machine learningPCAKNN / SVM
Completed team project

Mobile manipulator

Built a line-following mobile robot that could draw on a whiteboard, improving the design through repeated hardware and sensor tests.

ArduinoMecanum wheelsElectronics
Student-led project

SheffHEPP · Control Systems Lead

Led the control-systems work for a student team designing and testing a small hydroelectric generator.

LeadershipMATLAB / SimulinkControls
Second-place team project

Engineering You’re Hired!

In a week-long team project, we explored uses for swarm robotics. Our idea was to use drones with UV sensors to identify affected crops and spray only the plants that needed treatment.

Agricultural dronesSensorsTeamwork
Independent creative project

Game Development

An original Minecraft Bedrock addon built around interconnected creature behaviours, unusual environments, custom tools and world-generation systems.

JavaScriptSystems design3D assets
WIP Projects

Autonomous Robot Vacuum Cleaner

A personal project to build a custom autonomous vacuum cleaner that can navigate around obstacles.

ROS 2Multi-agent AISimulation
Planned experiment

Creatures that learn to walk

Physics-driven virtual creatures that discover locomotion strategies through reinforcement learning, evolutionary challenges and changing terrain.

MuJoCoRLPyTorch
Planned experiment

Artificial life sandbox

A self-contained ecosystem of autonomous creatures that compete, adapt and evolve as their environment changes.

Neural networksGenetic algorithms
// About me

I enjoy seeing how things work,then trying it myself.

I'm a mechatronics and robotics engineering graduate from the University of Sheffield. My interests span autonomous systems, robot manipulation, embedded engineering, computer vision and machine learning.

What connects my work is curiosity about systems: how individual parts behave, how they interact, and how you can make the whole thing more intelligent, capable or surprising.

Outside university, I enjoy creative projects, from building Minecraft Bedrock add-ons with custom behaviours and 3D models to experimenting with smaller robotics projects.

RoboticsControl, simulation, integration
AI / MLPrediction, perception, learning
HardwareEmbedded systems, sensors, CAD
CreativeGame mechanics, world generation
// Technical toolkit

Things I work with

PythonC++CJavaScriptMATLAB / SimulinkIsaac Sim / Isaac LabROS conceptsComputer visionMachine learningEmbedded systemsCADLinuxGit
// Experience
Industrial robotics

Hardware Development Engineer Intern — Amazon

Ten months spanning hardware concepts, CAD, technical documentation, site deployment, validation and cross-team delivery.

Academic research

Robotic slip-detection dissertation

Built a simulation-to-real pipeline and LSTM that detected up to 79% of slip events within a four-second warning window.

University engineering

Control systems & machine learning

Completed individual projects in physical-system control and chemical biodegradability classification.

Team engineering

Leadership, robotics & design

Led control-system work for SheffHEPP and contributed to mobile-manipulation and swarm-robotics team projects.

Independent development

Bedrock Fieldnotes

Original addon systems, scripting, modelling, animations and environment design.

// Start a conversation

Let's build something interesting.

Interested in robotics, intelligent systems or creative engineering? I'm always interested in thoughtful projects and collaborations.

EMAILAdd email addressContact me directly ↗GITHUBAdd GitHub linkExplore my code ↗LINKEDINAdd LinkedIn linkConnect with me ↗CREATIVE WORKBedrock FieldnotesExplore the portfolio ↗
Abhinav / Project archive / 001
Final year research · Completed

Predicting grasp slip before it happens.

My final-year project developed a Real2Sim and Sim2Real pipeline for robotic manipulation, using synthetic data to train a model that could recognise the warning signs of grasp slip before an object was dropped.

ISAAC SIMPYTHONLSTMSIM2REALROBOT MANIPULATION
GRASP MONITOR / RESEARCH SIGNALS / GRIP / PREDICTION
01 /

The idea behind the project

Using simulation to help robots recognise an unstable grasp.

A robot can drop an object when its grip starts to slip. I wanted to see whether an AI model could recognise the warning signs early enough to predict a failure before it happens.

/ WHY SIMULATION?

I recreated the real robot in Isaac Sim, allowing it to perform many repeatable pick-and-place trials and collect training data without running every experiment on physical hardware.

/ WHY AI?

The warning signs of slip can be subtle. An LSTM model learns patterns in the robot and object sensor data so it can recognise when a grasp is becoming unstable.

02 /

How it works

From a simulated robot to predictions on real-world data.

01 // SIMULATE

Recreate the robot

Build a digital twin in Isaac Sim and automate pick-and-place experiments.

02 // COLLECT

Record the signals

Collect object movement, robot motion and sensor data from successful and failed grasps.

03 // PREDICT

Train an LSTM

The model looks at sequences of signals over time to detect patterns that appear before an object slips.

04 // TEST

Use real robot data

Test the simulation-trained model on physical robot data to see whether those patterns transfer to the real world.

Good predictions on real robot data are important because they show that the simulated experiments generated useful training data, not just results that work inside the simulation.

03 /

See it in action

Simulation footage, robot tests and visualisations from the project.

Isaac Sim — grasping experiments

A short look at the simulated robot generating pick-and-place data.

Watch on YouTube ↗
▶SLIP EVENTPHYSICAL ROBOT · 16:9
Slip event

The robot initially grasps and lifts an object, but the object begins sliding out of its grip. The aim is to detect the warning signs before it falls.

▶MISSED PICK EVENTPHYSICAL ROBOT · 16:9
Missed pick event

The robot attempts to grasp an object but does not secure it properly, so the pick fails rather than the object slipping from an established grasp.

+

Images, plots & diagrams

Supporting views of the experimental setup, data pipeline and results.

Data Pipeline Setup

The experimental setup links the simulation, data collection and slip prediction system.

Visualised Results

This chart shows the time difference between predicted and actual slip events. The peak near zero means many predictions were close to the moment the object slipped.

04 /

Results & what I learned

What the experiments explored, the technical lessons and next steps.

How early could the model predict a slip?

The buffer is the time window before a slip during which a warning counts as a detection. A longer window catches more slips. The last column shows how early successful warnings occurred on average.

Buffer (seconds)Slip detection rateAverage early prediction (seconds)
0.0018.15%0.00
0.2026.76%0.03
0.4035.28%0.10
1.0052.25%0.28
2.0067.96%0.57
4.0079.12%0.89

These results show how detection changes with the allowed warning window. The highest detection rate was 79% with a four-second buffer.

01 / MODEL PERFORMANCE

Early slip prediction

The LSTM detected up to 79% of slip events when evaluated with a four-second warning window, showing that simulated data can help predict real grasp failures.

02 / VALIDATION

Sim2Real evaluation

I validated the simulation-trained model against real-world robot data and used physical trials to assess where synthetic sensing transferred well and where the reality gap remained.

03 / FUTURE WORK

Closing the loop

A natural next step would be to connect the early warning to an adaptive controller that changes grip force or motion before an object is lost.

/BIGGEST CHALLENGE

The hardest part was transferring a model trained on clean, repeatable simulation data to noisier physical sensing. This required careful experimental design and consistent signal collection across both environments.

/REFLECTION

The project showed me how simulation, sensing, data engineering and machine learning must be designed as one system. Reliable results depended as much on the experiment and data pipeline as on the model itself.

Abhinav · Robotics & intelligent systems · Final year project
Abhinav / Experience archive / 002
Industry experience ·
AMAZON · HARDWARE DEVELOPMENT

From development to deployment.

During a 10-month hardware development internship at Amazon, I worked across early design, technical documentation, site deployment and pilot delivery. It taught me how engineering ideas have to work reliably beyond the prototype, in real operations.

HARDWARE DEVELOPMENTSOLIDWORKS CADSITE DEPLOYMENTPROJECT DELIVERY
FIELD ENGINEERING / WORKFLOW
DESIGN→TEST→DEPLOY
CAD / INTEGRATION / TROUBLESHOOTING
INDUSTRIAL ROBOTICS TEAMHARDWARE DEVELOPMENT EXPERIENCE
Team photo · Barcelona Team Event
01 /

The bigger picture

Why the work mattered and where I contributed.

My work covered the development lifecycle from initial concepts and prototypes through to deployment and validation. It required balancing technical performance with safety, site requirements, timelines and clear communication across engineering and operational teams.

/ MY ROLE

I led an individual technology-development project while also coordinating deployment and pilot activity. My responsibilities included CAD, high-level technical documentation, milestone planning, validation and stakeholder updates.

/ PROJECT CONTEXT

The work involved robotics upgrades at several UK sites and collaboration with engineering teams in Europe and North America. I have kept some details general for confidentiality.

02 /

What I worked on

Design, testing, integration and fault investigation.

01 // DEVELOP

Concepts & prototypes

I produced initial concepts and SolidWorks designs, then documented the technology at a level that supported review and continued development.

02 // COORDINATE

Site deployment

I coordinated onsite teams and contractors around daily deliverables, keeping work aligned with the deployment plan and site requirements.

03 // VALIDATE

Technical verification

After installation, I helped validate that upgrades met the required operational and technical criteria before completion.

04 // DELIVER

Pilot & stakeholder alignment

I supported a pilot through data review, milestone planning and weekly updates with teams across Europe and North America.

/ OWNERSHIP

The internship gave me responsibility for moving work forward independently: defining the next technical step, maintaining documentation and communicating progress when several teams depended on the outcome.

/ COLLABORATION

I learned to adapt technical detail to the audience, from practical onsite instructions to project updates and evidence used to support broader delivery decisions.

03 /

Areas of experience

The recurring themes across my internship.

Design & documentation

Turning early ideas into CAD concepts, prototypes and clear technical material that other engineers could review and use.

Deployment & validation

Planning installation activity, coordinating onsite delivery and verifying that completed work met defined requirements.

Cross-functional delivery

Using data, milestones and regular updates to keep technical and operational teams aligned across regions.

04 /

What I took away

The engineering and delivery lessons that stayed with me.

/ END-TO-END ENGINEERING

The experience strengthened my understanding of how design, documentation, testing and deployment fit together. A technically sound idea only creates value when it can be implemented and verified reliably.

/ PROFESSIONAL GROWTH

My biggest takeaway was the importance of ownership and communication. I became more confident planning work, coordinating people and presenting technical evidence clearly enough to support decisions.

PROJECT COLLABORATIONENGINEERING AND DEPLOYMENT TEAM
Team photo · Industrial robotics experience

Abhinav · Robotics & intelligent systems · Industry experience
Abhinav / Project archive / 003
Independent creative project · Game development

Bedrock Fieldnotes.

Explore the full addon portfolio ↗
/ WHY I WANTED TO DO IT

I chose this project to push my creativity and build something for a real audience. Unlike many of my university and hobby projects, a public release needs to feel polished and offer something different enough for players to choose it.

/ WHAT I LEARNED

Through this project, I developed skills in texturing, modelling, animation, JavaScript and JSON while managing a codebase containing thousands of files. More importantly, I learned to welcome criticism rather than become attached to my first idea. Regular playtesting with friends helped me identify problems, act on feedback and shape the addon around what players found fun and intuitive.

/ TAKEAWAYS

My biggest takeaway was that quality comes through iteration. Nearly every idea and asset changed and evolved during development, and comparing my work with established add-ons helped me identify weaknesses and refine the project to a more competitive standard.

// WHAT WENT INTO THE ADDON

More than just adding new mobs.

Bedrock Fieldnotes is a large, ongoing Minecraft Bedrock addon built around a post-apocalyptic, vanilla-friendly world. I designed its content to connect: exploring a biome introduces new creatures, materials and mechanics that give players reasons to venture further.

14custom biomes
35custom mobs
4boss encounters
01 / PROGRAMMING

Behaviour & game systems

I wrote JavaScript with the Bedrock Script API, along with JSON entity definitions, to create unique AI behaviours, interactions, progression and combat. Some systems respond to the player's actions or surroundings; others connect mobs, items and the world together.

02 / ENVIRONMENTS

World generation & exploration

I designed custom biomes, vegetation, structures, blocks and spawning rules, aiming for places that feel different to explore rather than just look different. World generation needed plenty of testing so features appeared naturally and worked alongside Minecraft's existing systems.

03 / ART & CONTENT

Models, textures & animations

I created my own blocky 3D creatures, textures, animations, items and visual effects. Each design needed to be recognisable, fit Minecraft's style and communicate its purpose in-game, while still working with the underlying code.

04 / DEVELOPMENT

Testing, balancing & iteration

Creating so much interconnected content meant repeatedly fixing bugs, adjusting mob behaviour, balancing rewards and testing with other players. I also built a separate portfolio and wiki to document the world and explain its mechanics to people seeing it for the first time.

Tools & skills used

JAVASCRIPTBEDROCK SCRIPT APIJSONBLOCKBENCH / 3D MODELLINGTEXTURINGANIMATIONWORLD GENERATIONDEBUGGINGPLAYTESTINGDOCUMENTATION

The dedicated Bedrock Fieldnotes portfolio goes much deeper into the world, its creatures and how everything fits together.

Explore Bedrock Fieldnotes ↗

Abhinav · Independent creative development
Abhinav / Project archive / 004
University of Sheffield · Individual project

Controlling a dual-rotor helicopter.

I modelled a dual-rotor helicopter and designed a controller that could follow a predetermined path while maintaining smooth elevation, pitch and travel on the physical system.

MATLAB / SIMULINKSTATE SPACELQRKALMAN FILTER
CONTROL SYSTEM / PHYSICAL PLANT REFERENCE / RESPONSE MODEL / ESTIMATE / CONTROL / VALIDATE
01 /

The control challenge

Turning a coupled physical system into predictable motion.

The helicopter's elevation, pitch and travel influence one another, so a command in one direction can disturb the other axes. The task was to model those interactions and create a controller that could track a path smoothly without sustained oscillation.

/ OBJECTIVE

Move the physical helicopter along a predefined trajectory while keeping its elevation, pitch and travel stable, responsive and within the project requirements.

/ MY RESPONSIBILITY

As an individual project, I owned the workflow from deriving the model and selecting the control architecture through to implementation, parameter tuning and physical validation.

02 /

How I approached it

Model, estimate, control and refine.

01 // MODEL

Derive the dynamics

I derived the governing equations and expressed the coupled helicopter dynamics as a state-space model.

02 // ESTIMATE

Filter sensor noise

A Kalman filter estimated the system state from noisy measurements so the controller could respond more consistently.

03 // CONTROL

Track the reference

I combined an LQR controller with feedforward gain to balance stability while reducing the error between target and current position.

04 // TUNE

Refine on hardware

I observed faults in the physical response and adjusted LQR and Kalman parameters by penalising the states most responsible for each error.

/ ENGINEERING DECISION

State feedback made it possible to balance several competing behaviours rather than tuning each axis in isolation. The cost matrix gave me a structured way to trade response speed against overshoot and control effort.

/ ITERATION

Simulation provided a starting point, but the final behaviour depended on repeated physical testing. Each fault in the response was traced back to the model, estimator or state weighting before the next test.

03 /

System demonstration

A portrait-format demonstration of the physical rapid-control system.

▶PHYSICAL SYSTEM DEMONSTRATIONDUAL-ROTOR HELICOPTER · 9:16 PORTRAIT
Path tracking on the physical system

The demonstration shows the helicopter following its reference path while controlling elevation, pitch and travel.

04 /

Results & key takeaways

What the final system demonstrated.

These figures compare the requested reference motion with the measured response of the physical helicopter. Together, they show how accurately the controller followed the planned trajectory and make delay, overshoot and steady-state tracking error easier to assess.

▧ELEVATION RESPONSEREFERENCE AND REAL ELEVATION
Elevation tracking response

This figure compares the target elevation with the helicopter's measured elevation. The separation between the curves reveals the controller's response time, overshoot and remaining tracking error as the system changes height.

▧TRAVEL RESPONSEREFERENCE AND REAL TRAVEL
Travel tracking response

This figure shows how closely the measured travel angle follows its reference path. It also demonstrates how the controller manages the coupled motion while the helicopter rotates through the required trajectory.

01 / PERFORMANCE

Path tracking

The physical system met the project requirements and followed its prescribed path while maintaining smooth motion across elevation, pitch and travel.

02 / STABILITY

Controlled response

Careful estimator and controller tuning reduced tracking error without introducing sustained oscillations into the response.

03 / TAKEAWAY

Models need evidence

The project strengthened my ability to connect mathematical modelling with practical tuning and to use observed system behaviour to guide each engineering change.

Abhinav · Robotics & intelligent systems · Rapid control prototyping
Abhinav / Project archive / 005
University of Sheffield · Individual project

Predicting chemical biodegradability.

I built and compared binary classification models that predicted whether a chemical was biodegradable from its measured features, achieving up to 89% accuracy.

DATA PREPARATIONPCAKNNSVMMODEL EVALUATION
BINARY CLASSIFIER / ANALYSIS FEATURE SPACE / DECISION BOUNDARY PREPARE / REDUCE / TRAIN / COMPARE
01 /

The prediction task

A practical comparison of classification approaches.

The project used chemical data to predict whether each material was biodegradable. I focused on more than accuracy: I built a clear, repeatable process and examined why one method fitted the data better than another.

/ OBJECTIVE

Clean a dataset with many features and build a model that could separate biodegradable chemicals from non-biodegradable ones.

/ MY RESPONSIBILITY

This was an individual project, so I completed every stage: cleaning the data, reducing the number of features, building the models and checking the results.

02 /

From raw data to comparison

A clear, repeatable machine-learning process.

01 // CLEAN

Prepare the data

I removed outliers and normalised the feature values so that scale differences did not dominate the models.

02 // REDUCE

Apply PCA

Principal Component Analysis reduced the number of features while keeping the most useful information.

03 // TRAIN

Build two classifiers

I trained K-Nearest Neighbours and Support Vector Machine models on the same prepared dataset.

04 // EVALUATE

Compare behaviour

I compared the results and looked at each model's strengths and limits on a fairly small dataset.

/ MODEL SELECTION

KNN worked better than SVM for this dataset. I linked that result to the data size, structure and preparation choices rather than assuming KNN is always better.

/ CRITICAL ANALYSIS

Overall accuracy is only one part of the result. A model should also be judged by its errors, the data available and how well it may work on new data.

03 /

Comparing model behaviour

From the reference distribution to class-by-class evaluation.

I used the same PCA projection and evaluation structure for both classifiers so that their behaviour could be compared consistently. The purpose of these plots was not simply to present a final score, but to understand where each model reproduced the underlying data structure and where its errors occurred.

Prediction distributions

REFERENCE → SVM → KNN
▧REFERENCE DISTRIBUTIONTRUE LABEL DISTRIBUTION
True data distribution

I projected the prepared dataset onto its first two principal components to establish a consistent visual reference. This taught me to use dimensionality reduction as an inspection tool while recognising that a two-dimensional plot cannot represent every feature used by the models.

▧SVM RESULTSSVM PREDICTION DISTRIBUTION
SVM prediction distribution

I plotted the SVM predictions in the same PCA space so they could be compared directly with the true labels. This helped me see how an overall accuracy value can hide where a classifier favours one class or struggles in overlapping regions.

▧KNN RESULTSKNN PREDICTION DISTRIBUTION
KNN prediction distribution

I repeated the visualisation for KNN using the same coordinates and class colours. Comparing the plots showed me how two models trained on identical prepared data can form noticeably different local decision patterns.

Confusion-matrix evaluation

ERROR TYPE / CLASS BALANCE
▧SVM ERROR BREAKDOWNSVM CONFUSION MATRIX
SVM confusion matrix

I used the confusion matrix to separate correct classifications from each type of error. It reinforced that model assessment should consider class balance and the practical meaning of false positives and false negatives rather than relying on accuracy alone.

▧KNN ERROR BREAKDOWNKNN CONFUSION MATRIX
KNN confusion matrix

Applying the same evaluation to KNN gave me a fair basis for comparison. I learned to judge a model by the pattern of its mistakes and its suitability for the dataset, not simply by which method appeared more sophisticated.

04 /

Result & key takeaways

What the project demonstrated beyond the score.

01 / PERFORMANCE

Up to 89% accuracy

The best-performing pipeline classified chemical biodegradability with accuracy of up to 89% on the project dataset.

02 / COMPARISON

KNN performed best

KNN outperformed SVM in this case, illustrating the importance of selecting a method that fits the scale and structure of the available data.

03 / TAKEAWAY

Data shapes the model

The strongest lesson was that cleaning, scaling and feature reduction can influence the outcome as much as the choice of classifier.

Abhinav · Robotics & intelligent systems · Machine learning
Abhinav / Project archive / 006
University of Sheffield · Team project

Designing a mobile manipulator that could adapt.

Our team designed and implemented an Arduino-based robot that followed a marked track and used its manipulator to draw on a whiteboard.

ARDUINOROBOT IMPLEMENTATIONMECANUM WHEELSELECTRONICS
MOBILE MANIPULATOR / BUILD SENSE / MOVE / DRAW / REDESIGN
SYSTEMLine-following mobile robot
OUTPUTWhiteboard drawing
APPROACHTest-led redesign
// The build story

Testing changed the design.

The first wheel position limited movement, the line-following system needed more work, and the original sensor could not track the line reliably.

I helped find the causes and redesign the robot with a different sensor, updated code and a more practical layout.

/ DOCUMENTATION

I also helped write the technical report, explaining our choice of mecanum wheels, sensor and sensor position.

Find the problem

Testing showed that the wheel layout limited movement and the first sensor could not follow the line reliably.

Find the cause

We separated the mechanical and sensor problems, then chose what needed to change.

Rebuild and test

A new sensor, updated code and a revised mechanism improved the line-following system.

VIDEO /

Mobile manipulator demonstration

Footage of the robot following the track and drawing on the board.

▶NAVIGATION AND DRAWINGMOBILE MANIPULATOR · 9:16 PORTRAIT
Navigation and drawing demonstration

The demonstration follows the physical robot as it tracks the route and uses its manipulator to draw on the whiteboard.

Abhinav · Robotics & intelligent systems · Mobile manipulator
Abhinav / Project archive / 007
Student-led project · Control Systems Lead

Leading the controls work for a hydroelectric generator.

SheffHEPP brought together a student team from different engineering fields to design a working hydroelectric generator for the village of Honley in West Yorkshire.

TEAM LEADERSHIPCONTROL SYSTEMSMATLAB / SIMULINKEXPERIMENTAL TESTING
ROLE / LEADERSHIPCONTROL
SYSTEMS
LEADMULTIDISCIPLINARY TEAM
WEEKLY PLANNING · TECHNICAL DELIVERY
01

Set direction

I led a small controls team and made sure our work supported the wider project.

02

Build the simulation

I built the control-system simulation and improved my MATLAB and Simulink skills.

03

Design and test at scale

We built and tested a smaller model so we could measure the system before working at full scale.

04

Use the test results

I processed the test data and used it to guide the control plan for the full generator.

// What I learned

My first experience leading a major project.

This project gave me early leadership experience on a large team. I learned how to communicate with the managers of other teams, keep a major project organised, set clear weekly goals and make sure each group understood how its work fitted into the wider plan.

LEADERSHIPCOLLABORATIONSIMULATION DESIGNCONTROL SYSTEMSPLANNING
▧SHEFFHEPP PROJECT TEAMMULTIDISCIPLINARY COLLABORATION
Project team

The wider SheffHEPP team brought together several engineering areas around one shared project.

▧TEAMWORK IN PRACTICEPLANNING AND PROJECT DELIVERY
Site Visit

We visited the proposed generator site several times to keep our work grounded in the real setting.

Abhinav · Robotics & intelligent systems · SheffHEPP
Abhinav / Project archive / 008
University of Sheffield · Week-long competition

Targeted crop care using small drones.

In a small team, we designed a drone swarm concept to detect crops affected by disease or pests using UV sensors, then spray only the affected plants to reduce chemical use.

2NDCOMPETITION
RESULT
// The one-week sprint

Turn a clear idea into a workable plan in one week.

01 / UNDERSTAND

Define the problem

We looked at how crop disease and pests are found, and how spraying a whole field can waste chemicals, damage wildlife and impact plant health.

02 / DESIGN

Build the idea

We explored research into how affected and healthy plants can respond differently to light. Our concept used drones with UV sensors to identify crops that might need targeted treatment.

03 / PRESENT

Explain the plan

As a small team, we split the work, checked progress often and presented the idea and supporting research to the judges.

Outcome: our concept placed second. I learned how to organise work quickly, explain a technical idea clearly and combine input from people with different skills.

Abhinav · Robotics & intelligent systems · Engineering You’re Hired!

Project figure
Full size final year project figure