News

Unitree Go2 Reinforcement Learning

Unitree Go2 Gets Reinforcement Learning Upgrade

February 20, 2026

Unitree Robotics has released a firmware update for its Go2 quadruped that replaces the traditional gait controller with a reinforcement learning-based locomotion policy. Early tests show significant improvements in rough terrain handling, with the robot able to traverse loose rubble and steep inclines that previously required manual teleoperation. The update is available to all Go2 owners and represents one of the first over-the-air RL deployments on a consumer-grade legged robot. The new policy was trained entirely in simulation using domain randomization, then transferred to hardware without fine-tuning — a workflow that could become standard for future consumer robotics platforms.

Figure Humanoid BMW Partnership

Figure Partners with BMW for Humanoid Deployment

August 12, 2024

Figure AI has expanded its pilot program with BMW, deploying a fleet of Figure 02 humanoid robots at the automaker's Spartanburg, South Carolina plant. The robots are handling bin picking, part transfer, and quality inspection tasks on the body shop floor. BMW reports a 15% throughput increase in the pilot zones, with Figure 02 units operating alongside human workers during standard shifts. The partnership is the largest commercial humanoid deployment to date. Figure credits the integration of OpenAI's vision-language models for enabling rapid task programming — operators describe new tasks verbally and the robot translates instructions into action sequences without manual coding.

Open-Source Soft Gripper Design

New Open-Source Soft Gripper Design Released

March 9, 2022

Researchers at MIT's CSAIL have published an open-source design for a pneumatic soft gripper that can be 3D-printed on consumer-grade printers using standard TPU filament. The design uses a single air channel to produce a three-finger conforming grip capable of handling objects from 5mm to 120mm diameter. The project includes printable molds, assembly instructions, and a simple Arduino-based control board. Files are available on GitHub under an MIT license. The team reports the gripper can handle over 10,000 grasp cycles before showing signs of fatigue, making it viable for light industrial and educational use at a fraction of the cost of commercial alternatives.

Drone Swarm Record Breaking Display

Drone Swarm Breaks Record with 5,000 Synchronized Units

November 28, 2020

Chinese drone manufacturer EHang has set a new world record by coordinating 5,293 drones in a synchronized aerial display over Shenzhen. The swarm formed animated 3D structures including a walking humanoid robot and a swimming whale, with individual drone positioning accurate to within 10 centimeters. The display used a combination of RTK-GPS and peer-to-peer UWB communication to maintain formation precision. The performance lasted 18 minutes on a single battery charge. Beyond entertainment, EHang says the underlying swarm coordination stack is being adapted for agricultural surveying and disaster response, where hundreds of drones could autonomously map large areas in minutes.

Boston Dynamics Spot Inspection

Boston Dynamics Spot Gets Autonomous Inspection Certification

June 15, 2018

Boston Dynamics has received ATEX Zone 1 certification for a new variant of Spot designed for autonomous inspection in hazardous industrial environments. The certified unit can operate in areas with flammable gas or vapor without human supervision, a first for any legged robot. BP and Shell have both committed to deploying the certified Spot units across offshore platforms in the North Sea, replacing manual inspection routines that currently require shutdowns and scaffolding. The robot uses thermal imaging, acoustic sensors, and gas detection equipment mounted on its payload rail to perform inspections that previously took teams of engineers several days to complete.

Underwater Autonomous Robot Survey

Underwater Robot Completes 30-Day Autonomous Ocean Survey

January 22, 2016

The Monterey Bay Aquarium Research Institute (MBARI) has completed a 30-day autonomous survey using a fleet of three bio-inspired underwater robots. The manta ray-shaped gliders covered over 2,400 kilometers of ocean, mapping seafloor topography and collecting water quality data at depths up to 500 meters. The robots used wave energy harvesting to supplement their battery power, enabling the extended mission duration. The data has been shared with NOAA for integration into ocean health monitoring programs. Researchers noted that the bio-inspired propulsion produced virtually no acoustic disturbance, allowing the robots to collect marine mammal vocalizations that traditional propeller-driven AUVs would have masked.