Dexel Care · Scenario datasets for robotics teams

Care datasets for robots around people.

Dexel builds reviewed care scenarios for robots: when to help, wait, stop, and call a caregiver.

Bangalore · London · Dexel Care
Sample care case
Chair stand
Reviewed
SituationAn elderly person tries to stand from a chair.
RiskUnstable posture, reach for support, possible fall.
CorrectAsk first, stabilise only if needed.
EscalateConfusion, pain, repeated instability.
scroll
CHAIR STAND· FALL RISK· REFUSAL· PRIVACY· ESCALATION· RECOVERY· CHAIR STAND· FALL RISK· REFUSAL· PRIVACY· ESCALATION· RECOVERY·
The Problem
Robots need
to know when
not to act.

A care robot can copy a movement and still make the wrong decision.

The missing label is the care call: help, wait, stop, or escalate. Internet video, warehouse logs, and simulation do not provide that context.

🧠
When to help
A person reaching for a table may need support, privacy, or nothing at all. The useful label is the care decision, not just the movement.
🛡️
When to stop
Refusal, discomfort, confusion, and bathroom-adjacent assistance change what a robot should do. Dignity is part of the action space.
⚠️
When to call a caregiver
Near-falls, repeated instability, pain, and confusion are not edge cases. They are the moments a care robot must recognise first.
CARE Aging populations need
trustworthy assistance
RISK Human environments contain
unsafe edge cases
DIGNITY Assistance must preserve
privacy and autonomy
LABELS Care decisions become
training labels
Dexel starts with one hard question.
What should a robot do around a vulnerable person?
Dexel
Care.

Built around care moments where safety, consent, comfort, and escalation decide the right response.

01
Elderly Assistance
Support during standing, walking, sitting, reaching, and fatigue moments.
02
Fall-Risk Detection
Unstable posture, unsafe movement, proximity risk, and object hazards.
03
Patient Dignity
Privacy, tone, personal space, discomfort, and autonomy-preserving assistance.
04
Caregiver Escalation
When to continue, when to ask, when to stop, and when to call a human.
05
Confusion and Refusal
Hesitation, anxiety, refusal, misunderstanding, and recovery from wrong assumptions.
06
Personal Space
Safe approach, respectful distance, assisted movement, and comfort boundaries.
CARE ROBOTICS· ASSISTIVE SYSTEMS· ROBOTICS LABS· CARE PROVIDERS· HUMAN FACTORS TEAMS· SAFETY RESEARCH· CARE ROBOTICS· ASSISTIVE SYSTEMS· ROBOTICS LABS· CARE PROVIDERS· HUMAN FACTORS TEAMS· SAFETY RESEARCH·
What We Capture
What Dexel
Care captures.

Dexel Care captures the moments ordinary robot datasets miss: assistance, fall risk, dignity, refusal, escalation, and recovery.

First Dataset Vertical
Dexel Care.

For assistive robots that need to behave safely, respectfully, and consistently around people.

Each case names what helped, what risked harm, when to escalate, and what should never repeat.

Dataset Categories
Safety Decision
Fall risk, unsafe movement, hazards, escalation moments.
Dignity and Comfort
Privacy, personal space, tone, discomfort, autonomy.
Failure and Recovery
Spills, slips, refusal, confusion, near-falls, wrong assumptions.
Human Intent
Hesitation, fatigue, anxiety, reaching, searching, needing support.
Why Care First

Care makes the hard cases visible early: hesitation, refusal, privacy, fear, unstable movement, and the moment a human should take over.

— How It Works
Scenario Design

Care experts define the moment: chair stand, near-fall, refusal, privacy boundary, discomfort, or escalation.

Synchronized Capture

We record the scene, movement, speech, objects, distance, and caregiver response in controlled settings first.

Expert Labeling

Each scenario is labelled for intent, risk, correct action, wrong action, dignity impact, escalation threshold, and recovery path.

Dataset Output

Each case leaves with the video, sensor traces, labels, review notes, and recapture rules a robotics team can actually use.

Who We Build For
Built for
robots that
need to work
near people.
Robotics Labs
  • Robot learning researchers
  • Robot foundation model teams
  • Policy evaluation groups
  • Safety and alignment teams
  • World model researchers
Care Robotics
  • Assistive robotics companies
  • Elder-care robotics teams
  • Home assistance platforms
  • Human-robot interaction teams
  • Clinical automation researchers
Care Providers
  • Elder-care networks
  • Rehabilitation providers
  • Hospitals and clinics
  • Caregiver training teams
  • Human factors researchers
Why India
Why India
works.
01
Care demand and complexity
India combines family-led elderly care, dense urban homes, healthcare workforce depth, and varied care settings. That complexity creates better scenario coverage.
02
Diverse human environments
Homes, languages, norms, layouts, and caregiver behaviours vary sharply across India. Robots that only learn from narrow test settings will not generalize.
03
Operational data quality
This is not a low-cost labour argument. It is a data-quality argument: consented scenarios, trained care context, and repeatable operations built for scale.
Our Mission
Care data
from India,
built for
robotics teams.

Dexel Labs turns care scenarios into labelled datasets: when to help, when to wait, when to stop, and when to call a caregiver.

We start with care because bad robot behaviour is costly there: physically, emotionally, and legally.

The goal is simple: make care robots safer before they enter homes, clinics, and assisted-living environments.

NOW
Work With Us
Work on
Dexel Care.

Talk to the founding team about care scenarios, research partnerships, or joining the first dataset build.

Founded 2026 · Bangalore · London
About

India
building care
datasets for
robots around people.

Dexel captures care situations ordinary robot datasets miss: hesitation, refusal, privacy, unstable movement, escalation, and recovery.

The Founders
Three operators.
Hardware, data,
field systems.

Robotics companies are racing to build capable hardware, but safe human environments still lack the right datasets.

Internet video lacks physical context. Simulation misses discomfort and risk. Generic demos do not show when to wait, stop, or ask for help.

Dexel exists to build those missing datasets, starting with care.

Co-Founder Rikit Rathi

BEng Mechanical and Mechatronics Engineering (Hons), First Class — University of Hertfordshire. Four years at Cummins UK as a Program Manager — managing complex cross-functional engineering programs, coordinating across manufacturing, supply chain, and product teams.

Brings a systems-level understanding of how hardware programs scale, how physical operations are managed, and how to build tools that survive operational reality.

rathi.rikit@gmail.com
Co-Founder Shiven Raut

BTech and MTech in Environmental Engineering from IIT Bombay. Four years at Shell — working at the intersection of large-scale data systems, field operations, and deployment challenges.

Brings deep technical rigor, experience with industrial-scale data pipelines, and the ability to design systems that are robust enough to run in unpredictable field conditions.

shivenraut2000@gmail.com
Co-Founder Ajay Shriram

BEng Mechatronics Engineering — University of Hertfordshire. Published research on acoustic-based machine condition monitoring. Developed autonomous thermal test systems and seizure detection algorithms using ultrasonic data. Built physics-based fatigue life calculators for mechanical devices. Previously interned at LRDE, DRDO on CAD and FEA for a space-borne synthetic aperture radar.

The rare engineer who works across sensors, signal processing, embedded systems, and mechanical design simultaneously. GB patent filed as inventor in 2025.

ajayshriram104@gmail.com
Where We Are
Prototype,
then pilot.
2026
Founded around care scenario datasets.
Dexel Labs formed to capture the care decisions robots need to learn: help, wait, stop, escalate, and recover.
Now
Designing capture, annotation, and care protocols.
We are developing the synchronized capture stack, scenario taxonomy, annotation workflow, and first hiring plan for Dexel Care.
Next
First simulated care scenarios captured.
Run controlled scenarios around fall risk, confusion, discomfort, refusal, escalation, and recovery. Validate the scenario-to-dataset pipeline before field deployment.
2027
First Dexel Care dataset package.
Deliver reviewed care-scenario datasets to robotics, research, or care-automation partners for evaluation, safety review, and training.
India is
structurally
advantaged.

India offers a rare combination: care demand, trained healthcare talent, dense urban environments, family-led elderly care, linguistic diversity, and operational cost structures that make large-scale scenario collection possible.

This is not a low-cost labour argument. It is a diversity, complexity, and data-quality argument.

Robots that operate around people will need to generalize across homes, languages, norms, bodies, and care expectations. India gives Dexel a dense testing ground for that complexity.

0People · diverse care environments
0Official languages · multilingual care context
CAREFamily-led and professional care contexts
SCALEHigh-variation care scenarios for evaluation and training
What We Believe
The things we
keep coming
back to.
01
Robots need the right care call, not just demonstrations.
A motion trace can show what happened. Care labels explain what should have happened, what was unsafe, and why escalation mattered.
02
Care environments reveal the hardest human-robot problems first.
Care forces robots to interpret risk, discomfort, autonomy, refusal, and personal space. Those are the behaviours that decide trust.
03
Failure data is more valuable than perfect demos.
Near-falls, wrong assumptions, confusion, refusal, and recovery paths are not edge content. They are the dataset.
04
India is a structural advantage, not a cost advantage.
The value is variation: care settings, languages, home layouts, norms, and operational density. Data quality comes from complexity.
INDIA
Where We Go Next
Talk to us
about Dexel Care.

Partner with us on care scenarios, dataset design, capture tooling, or early hiring.

CARE
Join Dexel

Build the
Dexel Care
dataset.

We are hiring the first engineers, operators, and care-domain specialists building Dexel Care.

Rikit, Shiven and Ajay Co-Founders · Dexel Labs
What the work
requires.

You will work across field capture, data systems, annotation workflows, hardware integration, and dataset quality. The work is early, technical, and operational.

Scenario
Designing care situations worth capturing
Turning fall risk, confusion, refusal, privacy, and escalation into repeatable scenarios with clear consent, safety boundaries, and label definitions.
Capture
Making synchronized recording reliable
Testing video, audio, spatial, movement, object, and care-decision evidence against synchronization, privacy, and completeness requirements.
Annotation
Turning care decisions into labels
Building the workflow that labels intent, risk level, correct response, wrong response, dignity concern, escalation threshold, and recovery path.
Feedback
Turning model failures into new datasets
When a partner flags an unsafe or awkward robot behaviour, you map the missing signal, adjust the protocol, and package the next dataset around that failure case.
What You'll Build
Capture.
Annotation.
Dataset QA.

The first team will design the scenario protocols, capture stack, annotation tools, consent workflows, and quality framework behind Dexel Care.

This is systems work for robots operating around humans.

The output of your work should be usable by robotics teams, safety researchers, and care-automation companies without guesswork.

In your first 90 days:

Define scenario protocols for fall risk, discomfort, refusal, confusion, privacy, and escalation.

Build or operate the data stack that turns capture sessions into reviewed, reusable care cases.

Create quality checks for consent, synchronization, label consistency, and safety relevance.

Shape the first Dexel Care dataset before the process hardens into company habit.

What We Offer
Technical work.
Operational ownership.
🌍
Robotics Context
You will work on datasets intended for robots that interact with people, not abstract benchmarks.
🧠
Care-Decision Data
The core problem is not collecting more media. It is turning care decisions into labels robots can learn from.
🔧
Hardware + Software + Operations
The work spans capture devices, annotation systems, data pipelines, privacy workflows, and field operations.
📍
India as Data Advantage
The advantage is scenario diversity, care complexity, multilingual context, and dense operational learning.
Early System Design
You will influence the operating system of the company: what we capture, how we label it, and what quality means.
The right
operating
profile.
Less fit if
  • You need a mature playbook before starting
  • You prefer narrow ownership and fixed boundaries
  • You are uncomfortable with field constraints and messy data
  • You want to stay far from users, participants, or domain experts
  • You need all requirements to be written before acting
Best fit if
  • You have shipped systems people rely on
  • You can translate messy reality into clean systems
  • You care about safety, dignity, and consent as product requirements
  • You can work with engineers, care experts, and operators in the same week
  • You want India to build a globally relevant robotics company
Open Roles
Build the
first Dexel
Care team.

Show us what you have shipped. We care about systems, operations, research, and workflows you have made usable under real constraints.

Platform · Data Systems
Full-Stack / Data Platform Engineer
You build the internal platform that turns capture sessions into searchable, reviewable care-case records. This includes ingest, review tools, label workflows, QA checks, permissions, and export pipelines.
Show us what you've built
A data product, workflow tool, or platform system that people used. Include architecture, tradeoffs, and what broke in production.
Robotics Data
Computer Vision / Multimodal Data Engineer
You convert video, audio, spatial, movement, and object-interaction signals into clear evidence for care scenarios. You care about calibration, segmentation, confidence, occlusion, and data quality.
Show us what you've built
A computer vision or multimodal pipeline you shipped. Explain the input quality, failure modes, and how you measured usefulness.
Operations · Capture
Field Operations Lead
You run the operating system for scenario capture: participant scheduling, consent, equipment readiness, protocol adherence, site coordination, safety review, and session completeness.
Show us what you've built
An operation you built or improved under field constraints. Show the process, metrics, and the failure you learned from.
Domain · Care
Clinical / Care Domain Lead
You define care scenarios that are clinically sensible, ethically grounded, and operationally capturable. You help translate caregiver decisions into labels robots can learn from.
Show us what you've built
A care protocol, training program, clinical workflow, or safety process you designed or improved. Show how people used it.
Quality · Annotation
Annotation and Dataset Quality Lead
You own label quality, reviewer training, edge-case adjudication, taxonomy updates, audit trails, and dataset acceptance criteria. You make subjective care decisions consistent enough to train on.
Show us what you've built
A labeling workflow, QA framework, research coding system, or operational checklist that improved reliability.
Research · HRI
Human Factors Researcher
You study comfort, trust, autonomy, personal space, attention, fatigue, and refusal. You help Dexel define what respectful robot behaviour means in measurable scenario terms.
Show us what you've built
A study, product research program, HCI/HRI project, or behavioural framework that changed a design decision.
Privacy · Compliance
Privacy and Compliance Advisor
You help design consent, participant protection, data minimization, retention, access control, and audit processes for sensitive care-environment datasets.
Show us what you've built
A privacy, compliance, research ethics, or healthcare-data workflow you helped implement. Practical experience matters most.
APPLY
Show Us Your Work
Show us what
you've shipped.

Send a concise note with proof of work and the role or problem area you want to own.

Sample Scenario

A chair-stand
scenario,
turned into
data.

An elderly person tries to stand. Dexel records the scene, movement, spoken cues, risk, and care decision. Should the robot help, wait, stabilise, or call a caregiver?

Situation: chair stand
Risk: unstable posture
Correct: ask + stabilise
Wrong: pull abruptly
Escalate: confusion / pain
Recovery: stop + reassure
What gets
captured.
01 👁️
Scenario Video
Egocentric and room-level video capture the care interaction, participant movement, caregiver response, object context, and environmental constraints.
Views — participant + room context
Purpose — evidence for risk and response
Output — time-aligned scenario media
02 🔊
Audio and Speech
Conversation, reassurance, refusal, confusion, tone, and caregiver escalation cues are captured as part of the behaviour context.
Signals — requests, refusal, reassurance
Purpose — intent and comfort cues
Output — transcript + event alignment
03 🌍
Spatial Context
Room layout, chair position, walking path, obstacles, proximity, and object hazards are mapped against the care interaction.
Signals — layout, objects, distance
Purpose — safety and approach context
Output — spatial annotations
04 🦾
Movement and Interaction
Body pose, reaching, attempting to stand, hesitation, fatigue, contact, and object use are aligned to the scenario timeline.
Signals — posture, reach, support
Purpose — intent and risk inference
Output — movement event labels
05 ⚖️
Expert Review Labels
Care expertise turns media and sensor context into labels for risk, dignity, correct response, wrong response, escalation, and recovery.
Labels — safety, dignity, intent
Purpose — care decision labels
Output — care label set
Example:
standing from a chair.

A Dexel scenario is a bounded care situation with participant consent, safety controls, synchronized evidence, expert review, and a clear care-label checklist.

Scenario Elderly person attempts to stand from chair
Signals Video, audio, spatial layout, movement, object context, caregiver response
Controls Consent, safety supervision, scenario boundaries, privacy minimization, expert review
Output Care case with time-aligned evidence, labels, QA status, and export format
Why care
first.

Care scenarios expose the behaviours robots must get right before they can be trusted around people.

🛡️ Safety Decision

Fall risk, unsafe movement, unstable posture, proximity hazards, and moments that require escalation.

🧭 Intent and Comfort

Hesitation, fatigue, anxiety, reaching, searching, discomfort, refusal, and unspoken need for support.

⚠️ Failure and Recovery

Near-falls, wrong assumptions, confusion, refusal, spills, slips, recovery paths, and when to request human help.

The evidence
behind each label.
VISION ARRAY CHEST CAM HUB WRIST CAM GLOVE GLOVE Camera / active sensor IMU pod Glove sensors
Egocentric Video
Participant-perspective video shows approach, attention, reaching, support, and what the person can see during a care interaction.
Video
Room Video
Room-level context captures caregiver position, furniture, personal space, approach path, and hazards outside the participant view.
Context
Audio Context
Speech, tone, reassurance, refusal, confusion, and caregiver escalation are aligned to the scenario timeline.
Audio
Spatial Layout
Depth, room structure, chair placement, walking path, object hazards, and proximity constraints are converted into scenario context.
Spatial
Body / Pose
Posture, balance, reach, hesitation, fatigue, attempting to stand, and unstable movement are captured as risk and intent evidence.
Motion
Hand / Object Interaction
Contact, support, reaching, grabbing, stabilising, and object use are labelled against safety, comfort, and correct response.
Interaction
Expert Care Labels
Expert review adds human intent, risk level, safety signal, correct response, wrong response, dignity concern, escalation threshold, and recovery.
Labels
Labels need
the same timestamp.

A care scenario only becomes useful when the evidence lines up: what the person did, what was said, where the risk was, how the caregiver responded, and which care label applies.

Misalignment turns a safety label into noise. Dexel lines up video, audio, room context, and movement before review so every label refers to the same moment.

The goal is not more streams. The goal is a trustworthy scenario record that can support training, safety review, and failure analysis.

Scenario Timeline
Aligned media · events · labels · QA
Video media
Audio speech
Spatial layout
Movement events
Review labels

Every care case links evidence to the same scenario timeline: intent, risk, dignity, correct response, wrong response, escalation, and recovery.

Scenario-to-Dataset Pipeline
From scenario
to dataset.
1
Scenario Design
Care experts define the safety-critical moment, expected behaviours, risks, escalation criteria, and recovery paths.
2
Participant Consent
Consent, privacy scope, data minimization, supervision, and scenario boundaries are confirmed before capture begins.
3
Synchronized Capture
Video, audio, spatial layout, movement, object context, and caregiver response are captured against one scenario timeline.
4
Expert Annotation
Care and human-factors reviewers identify intent, risk, discomfort, correct response, wrong response, and escalation threshold.
5
Risk / Dignity Labels
Safety, privacy, autonomy, personal space, tone, and acceptability are encoded as explicit labels.
6
QA and Validation
Reviewers check evidence alignment, label consistency, consent status, scenario completeness, and edge-case coverage.
7
Dataset Packaging
Completed scenarios are packaged for training, policy evaluation, safety research, or partner export formats.
8
Failure Feedback
A robot or partner flags a failure case. Dexel maps the missing care decision, designs targeted recapture, and updates the dataset.
What a robotics
team receives.

Each dataset output preserves the evidence behind the label. A safety signal is linked to the video, audio, spatial context, and expert review that produced it.

Example record: elderly person attempts to stand from chair. Human intent: bathroom / mobility / discomfort. Risk level: medium-high. Safety signal: unstable posture. Correct response: verbal check + stabilise + call caregiver if needed. Wrong response: pull arm abruptly. Dignity concern: preserve autonomy, avoid panic. Escalation threshold: repeated instability / confusion / pain. Recovery: stop, reassure, request human help.

Scenario
Care case — situation description, participant context, environment context, safety setup, and capture metadata.
Default
Labels
Care label set — intent, risk, dignity, correct action, wrong action, escalation threshold, and recovery path.
Default
Media
Time-aligned evidence — video, audio, spatial context, movement events, and object-interaction markers.
Available
Export
Export formats — JSON, tabular labels, media references, and partner-specific exports.
Available
QA
Validation record — consent status, reviewer agreement, scenario completeness, label confidence, and known limitations.
Review
Why it gets
better with use.
01
Scenario Library
Each care scenario adds new examples of intent, risk, dignity, escalation, and recovery. The taxonomy becomes more valuable as coverage grows.
02
Annotation Discipline
Subjective care decisions have to become consistent enough to train on. Reviewer calibration, label definitions, and QA history compound over time.
03
Care Partner Protocols
Consent, privacy, scenario design, and safety review become an operating process. Trust and repeatability become a defensible asset.
04
Failure Feedback Loop
Every robot failure can become a targeted capture protocol. The dataset improves around the moments that matter most.
TECH
Bring Us A Failure Case
Turn it into
a capture protocol.

Talk to the founding team about care scenarios, label sets, research collaboration, or early technical roles.