Autonomous Driving

Civilisation spent a hundred years building itself around fossil fuel. Autonomy is how we start building it for something else.

We do not chase drive-anywhere. We engineer autonomy that runs and pays inside a known operating domain, on an open, auditable stack, with a human always one step away through teleoperation. Then we widen it.

Widen the domainSee the self-driving car
The autonomy computer and sensor electronics fitted in the test bed’s boot, Malaysia Autoshow 2023
First on public highways
2016Malaysia’s first self-driving car, and Southeast Asia’s first on public highways
Sensors fused
3+LiDAR, radar, camera, plus GNSS/IMU
Control loop
10–100 HzPlanning to control, on automotive compute
Localisation
cm-levelWith HD maps, even where GPS is weak

Domain first. Then wider.

REKA archive photograph of CRETA on a coned demonstration course at Malaysia Autoshow 2017; not a newly claimed Cyberjaya test.

Domain first

We bound the world the machine operates in. Not waiting for Level 5, we start where autonomy already works and pays: yards, ports, farms, fixed routes. Prove it inside a boundary, then widen it.

Open stack

We build on Autoware, auditable and inspectable, not a sealed black box. When a regulator, an insurer, or an operator asks how a decision was made, “the model decided” is not an answer. Every layer of our stack can be opened, examined, and replaced.

Human in the loop

Teleoperation is a designed layer, not a failure mode. The machine hands over cleanly at the edge of its domain. A person steps in, the situation is studied, the domain widens. A human is always one step away.

A decade in the field

We put Malaysia’s first self-driving car on the road in 2016, the first in Southeast Asia on public highways, before most of the region had a serious programme. That head start is the shape of the real problem understood from the inside, on our own roads, in our own conditions.

Widen the domain, watch the difficulty explode.

Every step from a fenced yard toward “anywhere, any weather, any time” multiplies the rare edge cases a system must handle safely. Drag the slider. Illustrative: the point is not the exact numbers, it is the shape. Difficulty is not linear in ambition.

AI-generated illustration of dense urban traffic in heavy tropical rain, showing difficult autonomous-driving conditions; not an actual REKA test route.
Operating domain
Fixed yard / port
1×relative long-tail difficulty
Ships today: REKA territory

Teleoperation plus a tight ODD. Fenced, mapped, predictable. This is where autonomy already pays, and where we deploy first.

The operating domain: in domain, at the boundary, out of domain Three zones. In domain the machine is autonomous, and every decision is logged and auditable. At the boundary, control transfers to a remote operator over the teleoperation layer. Out of domain, the machine comes to a safe stop. Prove it inside a boundary, then widen it. Out of domainsafe stop At the boundaryoperator assist In domainautonomous, logged, auditable Prove it inside a boundary, then widen it. teleoperation Remote operatora human is alwaysone step away The operating domain: in domain, at the boundary, out of domain Three zones. In domain the machine is autonomous, and every decision is logged and auditable. At the boundary, control transfers to a remote operator over the teleoperation layer. Out of domain, the machine comes to a safe stop. Prove it inside a boundary, then widen it. Out of domainsafe stop At the boundaryoperator assist In domainautonomous, logged, auditable Prove it inside a boundary,then widen it. teleoperation Remote operatora human is always one step away
Seth Linton, REKA's Systems Cartographer, scanning

SETHAutonomy is only trustworthy if it knows where it stops.

ReadoutHover or tap a zone to read what happens there.

Knows where it stops.

Autonomy is only trustworthy if it knows where it stops. The boundary is a designed feature: it is what makes the safety case provable. Pick a zone.

The full autonomy stack.

A layered stack, from the sensors bolted to the vehicle up to the operations that watch over a fleet. Each layer is a module we can deploy, tune, or replace independently, so you adopt only what your use case needs.

It is built on open standards: as a member of the Autoware Foundation, we tune the open self-driving stack for the region’s roads, weather and traffic, and feed improvements back upstream. No vendor lock-in, full auditability for safety cases, and a stack your own engineers can inspect.

REKA archive photograph of real roof-mounted autonomy sensors at Malaysia Autoshow 2023; hardware context, not a claimed calibration session.
The full autonomy stack Seven layers, from L7 Sensors and vehicle interface at the bottom, through perception and sensor fusion, localisation and HD mapping, prediction and planning (around 10 Hz), to control (around 50 to 100 Hz) and, on top, L1 Operations, tele-ops and safety. Sensing rises from the sensors to planning; control sends drive-by-wire commands back down to the vehicle interface. Each layer is a module that can be deployed, tuned or replaced independently. L1Operations, tele-ops & safetyfleet · teleoperation · safety case L2ControlMPC · PID · DBW~50–100 Hz L3Planningbehaviour · motion · lattice / OMPL~10 Hz L4Predictionintent · trajectory L5Localisation & HD mappingNDT · SLAM · GNSS/IMU · HD map L6Perception & sensor fusion3D detection · tracking · fusion L7Sensors & vehicle interfaceLiDAR · radar · camera · CAN / DBW The full autonomy stack Seven layers from L7 Sensors and vehicle interface at the bottom to L1 Operations, tele-ops and safety on top. Planning runs around 10 Hz, control around 50 to 100 Hz. L1Operations, tele-ops, safetyfleet · teleoperation · safety case L2ControlMPC · PID · DBW~50–100 Hz L3Planningbehaviour · motion · lattice / OMPL~10 Hz L4Predictionintent · trajectory L5Localisation & HD mappingNDT · SLAM · GNSS/IMU · HD map L6Perception & sensor fusion3D detection · tracking · fusion L7Sensors & vehicle interfaceLiDAR · radar · camera · CAN / DBW
Seth Linton, REKA’s Systems Cartographer, scanning

SETHEach layer is a module we can deploy, tune, or replace independently, so you adopt only what your use case needs.

ReadoutHover or tap a layer to read what it does.

The layers as a table
LayerNameWhat it doesTypical
L1Operations, tele-ops & safetyFleet monitoring, remote takeover, ODD enforcement and safety oversight.fleet · teleoperation · safety case
L2ControlTrajectory following and drive-by-wire commands, around 50 to 100 Hz.MPC · PID · DBW
L3PlanningBehaviour and motion planning in real traffic, around 10 Hz.behaviour · motion · lattice / OMPL
L4PredictionMulti-agent trajectory forecasting for vehicles, riders, pedestrians.intent · trajectory
L5Localisation & HD mappingcm-level pose from NDT scan-matching and GNSS-INS against an HD map.NDT · SLAM · GNSS/IMU · HD map
L6Perception & sensor fusionDetection, tracking and segmentation from fused LiDAR, radar and camera.3D detection · tracking · fusion
L7Sensors & vehicle interfaceLiDAR, radar, cameras, GNSS/IMU and the drive-by-wire interface.LiDAR · radar · camera · CAN / DBW
Sense Perceive Localise Predict Plan Act Planning ~10 Hz Control ~50–100 Hz

Sense, decide, act, every cycle.

The driving loop runs continuously on the vehicle. Each stage feeds the next, then the world is sensed again, dozens of times a second: planning around 10 Hz, control around 50 to 100 Hz, depending on platform and ODD.

SenseSynchronised LiDAR, radar, camera, GNSS/IMU streams, time-stamped on ROS 2.
PerceiveObjects, lanes and free space from fused sensors.
Localisecm-level pose on the HD map via NDT + GNSS-INS.
PredictWhere every nearby agent is likely to go next.
PlanA safe, comfortable trajectory through the scene.
ActSteering, throttle, brake via drive-by-wire.

Not a roadmap. A record.

REKA has been building autonomous technology in Malaysia and the region since 2016, before most programmes here existed. This work now lives in REKA’s Kinetik division. The full story is in the self-driving car case study.

REKA archive photograph from the 2016 CRETA Melaka–Kuala Lumpur highway drive; not a newly claimed Cyberjaya route.
AI-generated illustration of an autonomous mobile robot carrying totes along a marked warehouse lane; not a real REKA or customer deployment.
  1. 2016

    Malaysia’s first self-driving car

    CRETA MK I, a Proton Waja, drove from Melaka to Kuala Lumpur with no one in the driver’s seat: the first self-driving car in Southeast Asia on public highways.

  2. 2017

    MK II and MK III, and NICE Expo

    A Proton Inspira courtesy of MIDA, then two Proton Perdanas from NanoMalaysia and MARA, with LiDAR, radar and more cameras. Shown at NICE Expo, in space MARA and MARA Corp secured, where Malaysia’s Science Minister rode in the car.

  3. 2018

    Autonomous systems for a Singapore port

    An engagement on autonomous systems for port operations in Singapore.

  4. 2023

    A test bed with MARii

    Test runs on a dedicated route in Cyberjaya, then shown with MARii at the Malaysia Autoshow.

  5. 2024

    Autoware Foundation member

    Joined the Autoware Foundation as a member, after building on the open-source Autoware stack since 2021.

Now · shipping

Bounded domains

Autonomous machines in warehouses, yards, farms and fixed routes, paying for themselves today. Single sites. Single domains. Already running.

Near · 1–3 years

Coordinated fleets

Multiple machines running as one operation over secure networks, with teleoperation covering edge cases. The proven domain, now at fleet scale.

Far · the bet

The domain widens

As hardware, maps and regulation mature, the proven boundary expands. The organisations inside it first move first. We are getting clients ready.

Milestones, with sources.

The short version of the record. Each line says where it comes from: a dated source you can open, or REKA’s own record, with the documents to follow.

  1. 2016

    CRETA MK I, a retrofit kit on a Proton Waja

    One webcam, with motors, servos and REKA’s own electronics, including its RIG board, on the wheel and pedals. Driven from Melaka to Kuala Lumpur on the PLUS Highway by the end of the year, with no one in the driver’s seat. Bernama TV and Harian Metro covered it.

    Source: the CRETA case study and REKA’s press archive

    Documented
  2. 2017

    CRETA MK II and MK III

    MK II (mid-2017), a Proton Inspira courtesy of MIDA, with one puck LiDAR and two webcams. MK III (late 2017), two Proton Perdanas, a maroon one lent by NanoMalaysia Berhad and a white one from MARA, each with at least two puck LiDARs, two side radars and three cameras.

    Source: the CRETA case study and REKA press release, 16 October 2017

    REKA’s record
  3. Oct 2017

    NICE Expo 2017, with MARA and MARA Corp

    CRETA on show in Bukit Jalil, with MARA.

    Source: REKA’s press archive, October 2017

    REKA’s record
  4. Nov 2017

    Public rides at the Malaysia Autoshow, with MARii

    Visitors rode in the retrofitted car at the Malaysia Autoshow 2017.

    Source: press archive, Autoshow.com.my, 7 November 2017

    Documented
  5. May 2018

    Most research moves to software; autonomy carries on

    Regulation, infrastructure and investment, rather than the technology alone, were named as the things to address first.

    Source: REKA press release, 2 May 2018

    Documented
  6. Mar 2022

    MoU with Elmlab on autonomous-vehicle technology

    Co-development covering connectivity and teleoperation.

    Source: New Straits Times, March 2022

    Documented
  7. May 2023

    An autonomous test bed at the Malaysia Autoshow, with MARii

    Shown after test runs on a dedicated route in Cyberjaya.

    Source: REKA article, 19 May 2023

    Documented
  8. 2024

    Member of the Autoware Foundation

    After building on the open-source Autoware stack since 2021.

    Source: the Autoware Foundation’s announcement and REKA’s member profile

    Documented
  9. 4 Jul 2024

    A regional workshop on piloting self-driving cars in ASEAN

    In Yogyakarta, the co-founder took part by invitation, as a Malaysian industry expert and panel discussant, in a workshop on a regulatory pilot space to facilitate cross-border digital data flow for self-driving cars in ASEAN. It was organised by PUSTRAL UGM with an Indonesian ministry.

    Read: autonomous vehicle regulation in ASEAN

    REKA’s record

Documented: a dated source you can open. REKA’s record: confirmed by REKA, with the documents to follow.

Specifications.

Exact sensor suite, compute and ODD are scoped per platform and use case.

AI-generated illustration of a small autonomous shuttle on a bounded campus route; not a verified REKA deployment or evidence of driverless operation.
Core platform
Autoware (Universe / Core) on ROS 2
Perception sensors
LiDAR (mechanical / solid-state), automotive radar, mono/stereo cameras, GNSS/IMU
Localisation
NDT scan-matching · GNSS-INS fusion · HD vector + point-cloud maps · cm-level
Planning & control
Behaviour + motion planning · MPC / PID · drive-by-wire (CAN)
Compute
Automotive-grade GPU/SoC edge compute · real-time ROS 2 middleware
Operational design domain
Defined per deployment: shuttle routes, yards, ports, campuses, industrial sites
Validation
Scenario simulation · HIL/SIL · closed-course · supervised on-road
Operations
Teleoperation · fleet management · OTA updates · logging & replay