Robotics

Give a robot the sense to feel its own limits.

Proprioception for Physical AI

The Physics of Distribution Shift

Learned policies and vision-language-action models optimize for task completion across a training distribution. They operate in the semantic layer, predicting trajectories from visual tokens without a deterministic model of momentum, braking distance, or recovery reachability.

Physical deployments inevitably encounter structural distribution shift. When the workspace presents an unmapped obstacle, a shifted payload, or an unexpected human movement, the policy defaults to its statistical mean. The agent executes this generated trajectory directly into the physical infrastructure, translating a probabilistic error into a kinetic collision.

Hardware interlocks only trip after a physical boundary is breached. Between the learned policy and the safety PLC, the hardware attempts to execute a software hallucination at full mechanical torque.

A digital reflex arc for physical AI.

KAIROS routes structural margin back into the control loop before torque hits the joints, the way proprioception routes load feedback to the motor cortex. On every control cycle, the adapter evaluates the proposed motion envelope against stopping distance and workspace keep-out geometry, streaming Nominal, Caution, Restricted, or Hold directives to steer the planner clear, and gating the command into Pass, Constrain, Escalate, Mitigate, or Reject in under a millisecond.

Placement

KAIROS sits between the motion planner and the joint controller. Every candidate trajectory is evaluated against stopping distance and workspace geometry before torque is applied, while the certified safety PLC remains in place below the actuators as the hardware backstop.

Every trajectory must preserve a safe recovery path.

  1. Learned policy / VLAGenerates task-level intent and target poses from visual and semantic inputs.Safety today: statistical reward shaping
  2. Motion plannerComputes a candidate trajectory and velocity profile.
  3. KAIROSEvaluates the proposed motion envelope against braking distance, workspace keep-out locks, and safe-set reachability before actuation.PassConstrainEscalateMitigateReject↑ Continuous margin feedback to the planner: Nominal · Caution · Restricted · Hold.
  4. Joint controllerCloses the high-frequency torque and velocity loop at the actuators.
  5. Actuators & encodersExecute physical motion and stream joint-state telemetry.
  6. Safety PLC & interlocksCertified light curtains, force limits, and emergency stops.Safety today: trips after physical breachThe certified hardware backstop. KAIROS gates trajectories upstream so emergency stops remain a last resort.

Robotics lifecycle

One safety boundary from simulation to conformity file.

Four dedicated Rust crates bind the same structural-margin math across the engineering lifecycle: calibrating thresholds in simulation, shaping policies during RL training, gating trajectories at runtime, and sealing frozen-system validation audits.

01 / 04 · Sim-to-real parameter derivation

Calibrate

Every threshold is anchored to simulation evidence.

Today
Safety thresholds are hand-tuned on physical hardware by trial and error, leaving speed and clearance limits without a reproducible derivation.
With KAIROS
Parameters are derived against seeded scenario populations and ground-truth oracles, then locked into a signed, digest-pinned calibration package.
Owner
Safety and controls engineering

02 / 04 · Safe-by-design policy learning

Train

Policies learn inside their runtime boundary.

Today
The policy optimizes for task reward in simulation and collides with the runtime safety filter in deployment, causing oscillation and stalled tasks.
With KAIROS
Native Rust reward and counterfactual-margin bindings run inside the RL loop so the policy internalizes the exact boundary enforced at runtime.
Owner
ML and autonomy engineering

03 / 04 · Pre-actuation runtime gate

Deploy

Slow the trajectory before a hard stop is needed.

Today
Hardware interlocks operate as a binary tripwire after a keep-out plane is breached, halting the entire cell on every near-miss.
With KAIROS
Every proposed motion envelope is scored against braking distance and workspace locks before actuation, resolving near-boundary moves to a safe velocity scale.
Owner
Controls, autonomy, and safety engineering
Verdicts
Pass · Constrain (velocity-scaled) · Escalate · Mitigate · Reject, plus shadow-mode evaluation.

04 / 04 · Frozen-system conformity audit

Validate

Deterministic reproduction from committed transcripts.

Today
Validation relies on ad-hoc test runs that cannot be independently reproduced and silently invalidate whenever a model weight or threshold changes.
With KAIROS
A digest-pinned System Under Test manifest is scored in a single access-guarded look against geometrically disjoint held-out cohorts and adversarial scenario populations, producing Wilson-bounded competence and safety ledgers.
Owner
Functional-safety leads and conformity assessment
Evidence
Portable .krtb transcript bundles re-verify bit-for-bit via kairos robotics reproduce with no simulator installed.

SAF-001 · Runtime safety and operability

Zero collisions across 1,800 trajectories.

  • 0 / 1,800Collision hazards acceptedA position-only baseline gate evaluated on the exact same 1,800 trajectories accepted 1,200 collision hazards.
  • 81%Recoverable trajectories preservedResolved near-boundary motions through velocity-scaled Constrain verdicts with zero added collision risk.
  • 1 passNo-tuning frozen auditA digest-pinned System Under Test manifest scored once against a pre-registered cohort and replayed bit-for-bit.

A frozen build of the robotics gate, scored once against 1,800 MuJoCo scenarios, with the cohort, the baseline, and the statistical thresholds committed before the run.Read SAF-001

CMP-001 · Neural policy competence

Twenty-one of twenty-four sealed targets. Scored once.

  • 21 / 24Sealed held-out targets reachedCleared the pre-registered 0.50 Wilson lower bound at 0.725, compared to 0 / 24 for the untrained initialization.
  • 5 / 5Certified policy seeds passingEvery seed that passed the training gate cleared the held-out competence bar on the first and only look.
  • 1 lookAccess-guarded evaluationGeometrically disjoint held-out targets remained sealed behind a digest guard that halts on any unauthorized read.

A learned 6-DoF neural reach controller evaluated in a single access-guarded look against 24 geometrically disjoint held-out targets under a pre-registered Wilson lower bound.Read CMP-001

Compliance and regulation

Every robot ships with a black box. Now it needs a flight recorder.

Without KAIROS

The neural black box

Learned policies map sensor tokens to joint commands through millions of opaque weights. Post-incident logs show what the actuator did, leaving the decision boundary unrecorded and irreproducible.

With KAIROS

The pre-actuation flight recorder

Every proposed motion envelope, predicted stability margin, velocity-scale evaluation, and gate verdict is sealed into a SHA-256 hash chain that replays byte-for-byte from committed transcripts.

EU Machinery Regulation 2023/1230Applies 20 January 2027

Machinery with self-evolving behavior enters mandatory third-party conformity assessment. Notified bodies require a deterministic technical file; a statistical confidence interval is not a safety case.

Who relies on the record

  • Conformity AssessmentPortable .krtb evidence bundles re-verify every safety margin and gate verdict from committed bytes without requiring access to the simulator or policy weights.
  • Underwriting & LiabilityIncident transcripts reconstruct the exact proposed trajectory, the workspace lock state, and the mathematical reason a motion was passed, constrained, or rejected.
  • Technical DiligenceClosed-loop gate-on vs. gate-off ablation reports quantify containment and operability deltas across frozen scenario populations.

Scope a Robotics Evaluation

Scoped evaluations test the KAIROS robotics adapter directly against a team's own embodiment and workspace geometry. An engagement scopes a single control-loop decision point, ingests the robot's kinematic limits alongside a recorded trajectory cohort or MuJoCo/Isaac simulation scene, and runs a frozen validation pass.

Deliverables include a signed calibration artifact, a closed-loop gate-on vs. gate-off containment and operability report, and a portable .krtb replay bundle structured for EU Machinery Regulation 2023/1230 technical files. Full contribution scope, embodiment formats, and geometry redaction rules are published in the robotics partner data specification. Complete the form on the right to scope a run.

Request received. We'll be in touch.

Privacy Policy

Effective: October 2026

1. Data Controller

AnankeLabs AB (Ogan 12, 153 95 Järna, Stockholms län, Sweden) is the data controller for personal data submitted through anankelabs.io under the EU General Data Protection Regulation (GDPR).

2. Data Collected

AnankeLabs collects only the information voluntarily submitted through our contact, evaluation-scoping, and newsletter forms: name, email address, organization, role, domain of interest, and inquiry message. On-premise deployments of KAIROS Substrate operate offline and transmit no runtime telemetry to AnankeLabs. Design-partner calibration datasets are governed by a separate mutual NDA and local redaction protocol.

3. Purpose & Legal Basis

Submitted information is processed solely to respond to commercial, research, compliance, or investor inquiries (GDPR Art. 6(1)(b) and 6(1)(f)) or to deliver research and product dispatches following explicit subscription confirmation (GDPR Art. 6(1)(a)). AnankeLabs does not sell, rent, or broker personal data to third parties.

4. Cookies & Infrastructure Logs

This website uses no third-party advertising trackers or behavioral analytics cookies. Strictly necessary session cookies are used only where authenticated account features are enabled. Standard edge and server logs (IP address, request path, timestamp) are processed transiently for security, rate limiting, and abuse prevention.

5. Storage & Retention

Submitted records are stored on encrypted European and Cloudflare edge infrastructure. Inquiry records are retained only for the duration of the business relationship or evaluation, and newsletter subscriptions are retained until unsubscribed.

6. Data Subject Rights

Under the GDPR, individuals have the right to access, rectify, erase, restrict, or export their personal data, and to withdraw consent at any time. Requests are processed within 30 days. Individuals also have the right to lodge a complaint with the Swedish Authority for Privacy Protection (Integritetsskyddsmyndigheten, IMY).

7. Contact

For privacy requests or inquiries, email privacy@anankelabs.io.

Terms of Use

Effective: October 2026

1. Acceptance

Accessing and using anankelabs.io constitutes acceptance of these Terms of Use.

2. Intellectual Property

All software, documentation, research studies, calibration artifacts, and content on this site are the intellectual property of AnankeLabs AB. KAIROS Substrate, the KAIROS domain adapters (Robotics, AI Safety, and Cybersecurity), and the Spindle verification framework are proprietary technologies. No license to commercial binaries or patents is granted except under a written agreement or as explicitly stated in an open reproduction bundle.

3. Evaluations & Early Access

Scoped evaluations, partner calibration harnesses, and early-access software are provided under separate written evaluation or licensing agreements.

4. Integration & Safety Responsibility

Published studies, benchmarks, and documentation on this site are provided for technical informational purposes. Production deployment of KAIROS Substrate in physical machinery, autonomous agents, or security operations requires embodiment-specific calibration, system-level safety validation, and adherence to applicable statutory and hardware interlock requirements.

5. Limitation of Liability

This website and its public materials are provided "as is" without warranty of any kind. To the maximum extent permitted by law, AnankeLabs AB is not liable for indirect or consequential damages arising from the use of this website or reliance on its public informational materials.

6. Governing Law

These terms are governed by the laws of Sweden, with exclusive jurisdiction in the courts of Stockholm.

7. Contact

For legal inquiries, email legal@anankelabs.io.