Skip to content
Smriti AI

Automotive Intelligence

ADAS designed for the environments it must face.

Current businessBased on NVIDIA Hyperion

Smriti AI Labs designs, tests and refines ADAS solutions based on NVIDIA Hyperion for Indian automotive applications.

Discuss an automotive projectHow we engineer
Illustrative 3D render of a driving-assistance perception view on a left-hand-traffic road: the vehicle under evaluation with its planned path, a lead auto-rickshaw, a two-wheeler and a pedestrian flagged for attention, and an unmarked road edge.
Illustrative 3D render
Illustrative top-down drawing of mixed traffic on an Indian roadA two-way road with a footpath along one side and an unmarked, uneven edge along the other. Two-wheelers, three-wheelers, cars, a bus and pedestrians share the road with only a worn centre line and no lane lines. Abstract brackets mark road users inside an illustrative forward range of the vehicle under evaluation; one pedestrian near the unmarked edge is marked as uncertain, and a dotted line shows an estimated drivable edge. Objects outside the range carry no bracket. This is an illustration, not test data.

Key

  • Road user inside the illustrated range
  • Uncertain or partly visible
  • Estimated drivable edge
  • Illustrative forward range

In the scene

Two-wheelers, three-wheelers, cars, a bus and pedestrians share a two-way road with a worn centre line, no lane lines and an unmarked outer edge. Traffic keeps left; arrows show the direction of travel. Road users without a bracket sit outside the illustrated range.

DrawingMixed traffic, top-down view
TypeIllustrative graphic
NoteNot test data

Our work

Our work combines solution design, model adaptation, testing and evaluation.

Term · Intended operating conditions
The roads, traffic, lighting, weather and use cases a function is designed for. Behaviour is evaluated within them, and limitations are recorded against them.

We focus on understanding system behaviour within the intended operating conditions, identifying limitations and using evidence to guide refinement.

Indian roads ask a lot of driver assistance. Two-wheelers, three-wheelers, buses, cars and pedestrians share the same space, often without clear lane markings. Engineering for those conditions starts with describing them precisely.

Engineering loop

Design establishes the intent.Testing reveals the reality.Refinement connects the two.

Model adaptation sits between design and testing, fitting perception and related models to the environment. Each stage produces something the next can check, and the loop runs again with every iteration.

  1. 01

    Solution design

    Establishes the intent

    Functions and expected behaviour are defined against the intended operating conditions.

    Output

    Documented design intent

  2. 02

    Model adaptation

    Fits models to the environment

    Perception and related models are adapted to the roads, traffic and conditions the system will face.

    Output

    Adapted models and software

  3. 03

    Testing and evaluation

    Reveals the reality

    Structured evaluation of how the system actually behaves, including where it falls short.

    Output

    Evidence, including limitations

  4. 04

    Refinement

    Connects the two

    Evidence decides what changes next, and every change is evaluated again.

    Output

    A revised design intent

Refinement returns to solution design. Each iteration starts from the evidence of the last.

After refinement, the loop returns to solution design for the next iteration.

DrawingADAS engineering loop
TypeProcess
PracticeAutomotive

Ownership

What we engineer, and what NVIDIA provides

NVIDIA Hyperion, formerly NVIDIA DRIVE Hyperion, is NVIDIA’s compute and sensor reference architecture for vehicles. We did not create it. We design, adapt, test and refine ADAS solutions based on it, working with relevant NVIDIA DRIVE technologies where a solution needs them.

The platform provides compute and a reference architecture. How the solution behaves in its intended operating conditions is the engineering we do.

Technologies we build with

  • NVIDIA Hyperion
  • NVIDIA DRIVE

  1. Intended operating conditions

    Roads, traffic, lighting, weather and use cases, defined with the automotive team

    Smriti engineers

  2. ADAS solution design

    Functions and the behaviour expected of them within those conditions

    Smriti engineers

  3. AI software and model adaptation

    Perception and related models adapted to the target environment

    Smriti engineers

  4. Testing, evaluation and refinement

    Evidence of behaviour, including limitations, guiding each iteration

    Smriti engineers

  5. Programme requirements and data

    The automotive team's requirements, data and test environments, used with permission

    Integrated with permission

  6. NVIDIA DRIVE technologies

    Relevant NVIDIA DRIVE software and tools, used where a solution needs them

    NVIDIA provides

  7. NVIDIA Hyperion

    Compute and sensor reference architecture, designed by NVIDIA

    NVIDIA provides
  • Engineered by Smriti
  • Provided by NVIDIA
  • Customer systems and data, integrated with permission
DrawingADAS solution layers
TypeSystem diagram
PracticeAutomotive

Evidence

Evidence, not claims

We document how a system behaves within its intended operating conditions, and where it does not.

  • Conditions first

    Every observation is tied to the operating conditions, scenario and configuration it came from.

  • Limitations recorded

    Where the system falls short is documented as a finding, with the same care as what works.

  • Changes traced to evidence

    Each refinement points back to the findings that prompted it, and is evaluated again.

  • Published only when authorised

    Evaluation material appears on this site only with the authorisation of the programme it belongs to.

Evaluation record

Ref — — —

Function under evaluation
The ADAS function and the behaviour its design intends.
Intended operating conditions
Road types, traffic mix, lighting and weather the function is designed for.
Scenario
The situation evaluated, described precisely enough to repeat.
Configuration
Software and model versions, and the platform configuration used.
Expected behaviour
What the design intent says should happen.
Observed behaviour
What the system did, referenced to the evaluation material.
Limitation identified
Where behaviour fell short, or conditions the function does not cover.
Refinement decision
What changes next, and how it will be evaluated again.

Every finding carries its conditions. Limitations are recorded as findings, not left out.

DrawingEvaluation record structure
TypeReporting format
NoteNo test data

Engagement areas

Where we work with automotive teams

Engagements are scoped around a function, the conditions it must operate in and the programme's stage of development.

To start a conversation, tell us

  • The function you are developing
  • Its intended operating conditions
  • Where the programme is today
  • The evidence you already have
Discuss an automotive project

  • ADAS solution design

    Defining functions and their intended behaviour for Indian operating conditions, on architectures based on NVIDIA Hyperion and relevant NVIDIA DRIVE technologies.

  • AI software and model adaptation

    Adapting perception and related models to the roads, traffic and conditions a programme targets, and engineering the software around them.

  • Testing and evaluation

    Structured evaluation of system behaviour within the intended operating conditions, with limitations documented alongside results.

  • Evidence-led refinement

    Turning evaluation findings into prioritised changes, then evaluating again to confirm what each change did.

Long-term ambition

Long-Term Autonomy Research

Research toward Level 5 autonomy is a long-term ambition. It sits apart from the ADAS work we do today and is described here so the two are never confused.

Our current work is driver assistance: designed, tested and refined for the conditions a function is intended to operate in. It is not Level 5, and we do not present it as a step that has been reached on the way there.

Current business

ADAS design, testing and refinement

Driver assistance engineered for intended operating conditions, based on NVIDIA Hyperion.

Long-term research ambition

Research toward Level 5 autonomy

Not a current capability. It is not what our ADAS work delivers today.

DrawingResearch horizon
TypeIllustrative graphic
NoteUndated
Level 5 autonomyLong-term ambition
A research ambition within our automotive practice. It is not a current capability, and we make no claim to it.
Our current ADAS workCurrent business
Driver assistance designed, tested and refined for intended operating conditions. It is not Level 5 autonomy.
NVIDIA HyperionPlatform by NVIDIA
The reference architecture our solutions are based on. Using it does not by itself establish an automation level for any vehicle or for our work.

Automotive Intelligence

Bring us the function and the conditions it must face.

Tell us what you are developing, where it must operate and where the programme is today.
Discuss an automotive projectSee our portfolio