Updates on our stack, technical deep dives, and lessons from the field.
When a robotic arm approaches a singular configuration, task-space commands require infinite joint velocities. We explore Yoshikawa's manipulability index and explain why the safety layer must translate geometric limits into safe refusal.
In physical AI, alignment is not a preference, it is a boundary. We explore why safety cannot be implicitly learned and how Control Barrier Functions mathematically guarantee that a robot never violates its physical limits.
We explore the theoretical benefits of neuro-inspired spatial grids and why moving beyond standard Cartesian coordinate systems is essential for reducing distance calculation complexity in physical autonomy.
A visual system that only reacts to what it sees is insufficient for autonomy. We explore why robotics requires probabilistic world modeling and how simulating future states mathematically handles the problem of occlusion.
To avoid mechanical stress and rapid hardware degradation, an AI cannot simply connect dots. We explore the higher-order derivatives of physical motion and why deterministic interpolation must sit between intent and actuation.
To guarantee safe actuation, we must look beyond the weights of a neural network and return to the mathematics of spatial constraints. We explore why implicit learning is insufficient for physical autonomy and how differentiable optimization at the edge guarantees safe execution.
Real-world robotics is multi-modal. We explore why traditional regression fails in the face of non-uniqueness and how diffusion models provide a mathematical bridge between noise and physical intent.
Touch is no longer an auxiliary sense. It is the central bottleneck of general purpose physical intelligence. We explore how tactile intelligence is formalized through continuum mechanics, information theory, and control.
If 2024 and 2025 were about giving robots a brain through LLMs and VLMs, 2026 feels like the year we finally give them a functioning nervous system. We explore how physics, control theory, and modern machine learning merge in earnest.
Training a robotics foundation model is not an exercise in scaling parameters. It is an exercise in deciding what kind of world you want a machine to survive in. Unlike language or vision models, an RFM lives in time, friction, latency, contact, failure, and recovery.
We chose to open source part of Xolver not as a marketing gesture, but as an architectural decision. Closed systems create the illusion of progress. Open systems reveal where reality pushes back.
Xolver starts from a simple belief: Intelligence only matters when it survives contact with the real world. Our work begins where clean data ends and uncertainty begins.