Back to PsiBot
Psi SynRobot

Psi SynRobot

Psi SynRobot

Last updated: October 7, 2026
Locomotion Type
Wheeled
Stage
Mass Production
Height
180 cm
Degrees of Freedom
20 DOF
Weight
150 kg
Battery Life
8 h
Release Year
2025
Computing
NVIDIA Thor compute platform. The architecture is end to end: once an instruction is given, an upper vision-language model analyses the goods scattered on a table and plans the order in which to handle them, and a lower operation model breaks each item into sub-tasks and executes them in sequence. The model is trained on large-scale simulated data, chained across skills by a bidirectional training framework, and aligned with high-quality real-robot data to raise long-horizon success rates. Charging is by cable or automatic docking.
Sensors
Four RGB-D cameras and three fisheye cameras

Psi V1, which the company also calls Psi SynRobot and psi-SynRobot, is PsiBot's general-purpose robot platform for complex logistics, retail and industrial settings, and it went into volume production in April 2026. Its defining idea is that working and learning are the same activity: the machine is built around an operate-and-collect architecture in which every task it completes also deposits a training sample covering vision, touch and motion trajectory, so the fleet improves the model simply by doing its job and the company describes this as breaking the industry's question of whether data or intelligence comes first. It is built for continuous duty, designed to run stably around the clock with a full-dimension safety design, and it uses a multi-joint bionic configuration to handle long-horizon multi-step work with what the company says is industry-leading accuracy. A spatial perception system lets the same machine adapt to warehouse sorting, retail guidance and industrial assembly, which is how the company frames it as one machine for many uses. The published specification is a standing height under 180 cm, a single-arm reach of 683 mm, more than 8 hours of runtime, four RGB-D cameras and three fisheye cameras for perception, a waist with 2 degrees of freedom and a neck with 2 degrees of freedom, NVIDIA Thor compute, and charging either by cable or automatic docking. Its intelligence is end to end: an upper vision-language model looks at goods scattered on a table and plans the order to handle them, and a lower operation model decomposes each item into sub-tasks and executes them in sequence. The company reports that the model reaches a grasp success rate above 99 per cent from only twenty real-robot samples, and that it adapts to flexible objects strongly enough to recover and resume packing after being interrupted or disturbed.

Application Scenarios

Warehouse sorting and logistics; retail guidance and store operations; industrial assembly and production lines; long-horizon multi-step manipulation in unstructured environments; flexible-object handling; and continuous around-the-clock deployment as both an execution terminal and a data collection endpoint.

Subscribe to Weekly Report

Get weekly updates on humanoid robot funding, new companies, and product launches.