Development timeline
Capability timeline
Xolver's breakthrough Robotics Foundation Models (RFM) Model X1-D has been built with relentless pursuit and velocity, consistently evolving its architecture and achieving breakthroughs, for the physical AI future.
September
2024
The Foundation
- Established base autonomous scene generation system.
- AI reasoning steered by semantic cues.
- Optimized tokenization reduces action latency.
November
2024
Perception & Data Scaling
- Infrastructure for simultaneous multi-task training.
- Vision system prioritizes critical movement details.
- Accelerated training via human video demonstration.
- Improved precision via fine-grained visual reasoning.
January
2025
Intelligent Reasoning
- Look-ahead reasoning simulates possible futures.
March
2025
Efficiency & Stability
- Memory optimizations halved training costs.
- Dual-stream system for knowledge and skills.
- Eliminated jitter for smooth control frequency.
May
2025
Robust Planning & Agency
- Improved following by prioritizing text instructions.
- Agentic planner decomposes abstract requests.
July
2025
Physical Grounding & Diffusion
- 3D coordinates for accurate gripper control.
- Failure analysis refines internal logic.
- Diffusion-powered parallel action sequence planning.
September
2025
World Modeling & Imagination
- Imagined object states ensure sound plans.
- Unified architectures into world-modeling agent.
October
2025
Safety & High-Fidelity Simulation
- Safety layer blocks dangerous motor commands.
- Connecting AI to high-fidelity industrial simulators.
November
2025
Real-Time High-Frequency Control
- Massive models exert high-speed physical control.
- Self-calibration adjusts for hardware offsets.
December
2025
Advanced Logic & Verifiable Success
- Fewer thinking steps improve AI responsiveness.
- Geometry-grounded logic prevents physical violations.
- Dual-speed system for simple vs complex.
January
2026
Memory & Industry Standards
- Compressed history allows long-term context.
- Advanced trajectory blending solved robotic snap.
- Communications standardized for global industry benchmarks.
February
2026
Brain-Inspired Spatial Math
- Hexagonal coordinates reduce spatial math complexity.
- AI generates high-quality training data.
March
2026
Real-Time Reflection
- Text journal enables real-time behavioral adjustments.