KANGAROO

描述

KANGAROO is PAL Robotics' bipedal research humanoid, built in Barcelona for work on dynamic locomotion, reinforcement learning and embodied AI. It is sold as a development platform: customisable, instrumented and open enough to experiment on rather than a finished product. The engineering choice that defines it is the drive. Custom linear actuators with integrated force sensors sit close to the body, giving a wide range of motion with high speed and torque at the joints while keeping the legs light; decoupling leg length from hip swing improves efficiency further. The result is low moving inertia in the legs, which means better impact resilience and lower energy consumption, and it is what makes highly dynamic motion possible. Closed-loop force and torque controllers at joint level let researchers try their own control algorithms rather than only the supplied ones. It stands 1.60 m and weighs 50-65 kg depending on configuration, with 14 to 40 degrees of freedom in total: 6 per leg, 2 in the torso for yaw and roll, and arms in 4, 5 or 7 degree-of-freedom versions carrying 7, 5 or 3 kg fully extended and up to 30, 28 or 25 kg bimanually. Control runs on an EtherCAT loop at 2 kHz with 0.25-degree static and 0.5-degree dynamic accuracy, rated PL d to EN ISO 13849-1. Two Intel i7 computers with 32 GB RAM and a 1 TB SSD each, plus an optional NVIDIA Jetson GPU, run Ubuntu LTS with ROS 2 LTS and PAL OS, with access to actuator PIDs, a whole-body inverse-dynamics controller, visualisation and diagnostics tools and a VR teleoperation demo. Sensing includes four RGB-D cameras, 6-axis force/torque sensors and fingertip tactile pressure sensors, on a 976 Wh battery.

功能特色