GR-1 Humanoid Uses Multi-Camera Vision

Fourier Intelligence’s engineers opted for a vision-only approach with their sleek GR-1 general-purpose humanoid, foregoing a full suite of next-generation sensors like radar and LiDAR.

Six RGB cameras positioned around its frame equip the GR-1 with a complete 360-degree view of its surroundings.

This configuration supports the creation of a birds-eye-view map using camera data, facilitated by a neural network that learns from context to generate 3D spatial features and virtual objects.

Advanced Navigation Capabilities and Real-Time Detection Performance

The company explains that the technology converts data into a three-dimensional occupancy grid, enabling the GR-1 to navigate through both accessible and restricted areas. Recently, during outdoor walking trials, the robot reportedly showed “exceptional efficiency and accuracy in real-time detection of vehicles and pedestrians along sidewalks.”

Cost-Effective Vision System Enhances GR-1’s Environmental Perception

Like Tesla’s 2021 decision to prioritize vision systems for its Autopilot, this approach aims to substantially cut hardware costs while enhancing the GR-1’s environmental perception, ensuring safer and more precise operations.

The current GR-1 model differs greatly from the skeletal, open-faced biped prototypes introduced last year. According to Fourier’s product page, it can walk swiftly with a human-like gait across various surfaces, aided by adaptive balance algorithms that maintain stability on slopes.

The company states that the GR-1 boasts 54 degrees of freedom: three in the head and waist, seven in each arm, eleven in each five-digit hand, and six in each leg. The robot features a peak joint torque of 230 Nm. Additionally, it sports distinctive purple hip bumpers, adding a touch of style to its humanoid form.

The AI capabilities include a multimodal language model similar to ChatGPT, along with advanced semantic understanding, natural language processing, and logical reasoning.

Its previously empty head now features a high-definition display, audio speakers, and a microphone embedded in its face. The comprehensive vision system enables real-time mapping, navigation, obstacle detection, and collision avoidance.

“This advancement represents a new phase in our research on embodied AI,” commented Roger Cai, the company’s director of robot application research and development. “With our vision-focused solution, the GR-1 is positioned to play a crucial role in various applications such as medical rehabilitation, family services, reception and guidance, security inspection, emergency rescue, and industrial manufacturing.”

Read the original article on: New Atlas

Read more: Lenovo’s Six-Legged Robotic Dog Unmatched Agility