Develops a perception-based Safety Harness and world-model software for physical AI in the home robotics environment
Arkona Home Robotics
Berlin, GermanyNew on Amplifa
About Arkona Home Robotics
Arkona Home Robotics is a Berlin-based development company in the field of physical AI and home robotics. Its core product is a perception-grounded Safety Harness: a rule-based, blocking-by-default gate that runs in front of every executing instance (learned policy, planner, or world model). Every proposed action must first be confirmed as safe by perception; 31 preconditions are checked against a pre-simulated trajectory, including object and payload safety, human proximity thresholds referenced to ISO/TS 15066, sensor integrity, and robot self-limits. If a condition is not confirmed, the robot executes a predefined safe fallback. The Safety Harness is validated live in NVIDIA Isaac Sim and Isaac Lab, among others with the Franka Panda Arm; an adapter for the Unitree G1 is in early integration. In addition, Arkona conducts research on a predictive world model as well as multimodal perception, Instruction Grounding, and dexterous control. As a member of the NVIDIA Inception Program, the company is developing with the Arkona P-1 a research cell consisting of a 6-axis manipulator, overhead and wrist cameras, and an instrumented work surface.
- Site
- Berlin
- Processes
- Safety framework (Safety Harness) for physical AI, Validation in NVIDIA Isaac Sim / Isaac Lab, World model research, Robotics prototyping
- Materials
- Safety Harness, World Model, NVIDIA Isaac Sim / Isaac Lab, ISO/TS 15066
- Certificates
- Safety validation in NVIDIA Isaac Sim, Automated safety tests, ISO/TS 15066 reference check, ISO 9001
- Industries
- Home Robotics, Robotics Research
- 01
Perception (WorldState)
- 02
Prediction (Swept Path, Dynamics Adapter)
- 03
Verification (31 preconditions)
- 04
Decision (Permit or Block)
- 05
Execution or safe fallback
- 06
Logging and escalation
Software Components and Methods
Safety Harness
Rule-based, perception-grounded default-deny gate that checks every action for safety before execution.
World Model
Learned, predictive simulator of the robotics environment that plays through candidate actions before execution (research direction).
NVIDIA Isaac Sim / Isaac Lab
Simulation environment in which the Safety Harness is validated live.
ISO/TS 15066
Reference standard for human proximity thresholds within the safety checks.
Locations
Berlin
DEDevelopment and Research
- Materials
- Software for physical AI
- NVIDIA Isaac Sim / Isaac Lab
- Arkona P-1 research cell (6-axis manipulator, overhead and wrist cameras)
- Franka Panda Arm
- ANYmal-C Quadruped (validated)
- Unitree G1 (early integration)
Industries
Additional services
Safety Validation
The Safety Harness runs live in NVIDIA Isaac Sim and Isaac Lab with 308 automated safety tests and 24/31 live-confirmed hazards blocked; every decision is logged for audits. The checks include human proximity thresholds referenced to ISO/TS 15066 as well as robot self-limits.
| Evidence | Berlin |
|---|---|
| Safety validation in NVIDIA Isaac Sim | |
| Automated safety tests | |
| ISO/TS 15066 reference check | |
| ISO 9001 |
Development Status
Currently: two robots (Franka Panda Arm, ANYmal-C Quadruped) validated in NVIDIA Isaac Sim/Isaac Lab; Unitree G1 adapter in early integration; real-hardware validation and humanoid support on the roadmap