Specs
Last updated
Last updated
Height: 6'2" (188 cm)
Weight: 60 kg
Runtime: 5 hours on a 24V Li-Ion battery.
Body Material: A mix of 3D-printed ABS plastic and extruded aluminum. This keeps it lightweight but strong.
Onboard GPU: Nvidia Jetson Orin, capable of 275 trillion operations per second (TOPS), runs all the complex AI tasks Dropbear needs to function smoothly.
Dropbear can operate on its own using advanced AI systems:
Vision: runs finetuned SAM2 grounded models for real-time recognition of customized classes, allowing it to adapt to specific environments and objects. DeepSORT for tracking objects and maintaining spatial awareness.
Dropbear can be controlled remotely:
Network Streaming: Live video feeds from the robot’s cameras give operators a real-time view of its surroundings for better control.
Stereo Cameras: Dropbear has a 6-camera setup for depth perception, helping it gauge distances and interact with objects more accurately.
360° Camera: Provides a full view of its surroundings to improve spatial awareness.
Onboard Display: Shows a live simulation of what Dropbear is doing, which can be useful for troubleshooting and monitoring during operation.
Autonomous: Dropbear processes visual data to understand its environment and make decisions. It uses AI to navigate, manipulate objects, and adapt to changing conditions.
Teleoperation: Allows users to remotely control Dropbear with real-time feedback, making it feel like you’re right there with the robot.
Actuation: Electric actuators () power its movement. Electric systems are simpler to maintain and more energy-efficient than alternatives like hydraulics.
Reinforcement Learning: Reinforcement learning models trained on Open X-Embodiment and further finetuned using RT-X models for robot's decision-making capabilities, enabling it to perform complex manipulation tasks autonomously (). It learns and improves from its interactions, allowing it to perform more complex tasks over time, like manipulating small objects or navigating difficult environments.
Visual Language Models: Running an 8B vision-language model () captures detailed visual data, helping Dropbear respond dynamically to changes in the environment.
Audio: Uses , converting speech into text to understand and respond to spoken commands.
Remote Control: Operators can take control via , allowing precise operation in environments where autonomy might not be enough. This is useful for tasks requiring human judgment or delicate operations.
