Deep Robotics Lite3 Pro

The Deep Robotics Lite3 Pro is a research-grade quadruped platform that bridges exceptional locomotion capabilities with sophisticated AI perception. Featuring the proven RK3588 motion controller paired with an NVIDIA Jetson Xavier NX processor and Intel RealSense D435i depth camera, the Lite3 Pro delivers advanced perception without the complexity of full LiDAR integration. This configuration makes it ideal for researchers focusing on visual SLAM, neural network inference, deep learning-based perception, and AI-driven autonomous behaviors. The split-brain architecture keeps motion control independent while enabling full experimental freedom on the perception side.

The split-brain architecture with independent motion controller and perception processor enables researchers to modify AI algorithms without affecting the stability and reliability of locomotion control. This separation is critical for developing and testing new neural network models without risking robot stability.

The Intel RealSense D435i depth camera and NVIDIA Jetson Xavier NX provide professional-grade perception capabilities for visual SLAM, object detection, and semantic understanding tasks. This configuration optimizes cost-to-capability ratio for vision-intensive research applications.

• Dimensions: 610 × 370 × 445 mm (Standing)
• Weight: 12.2–13.5 kg (including battery, sensor package)
• Payload Capacity: 5 kg continuous; 7.5 kg maximum
• Battery Endurance: 1.5–2 hours continuous operation
• Operating Range: 4 km (with perception host)
• Max Speed: 4 m/s
• Slope Climbing: Up to 40°
• Stair Climbing: 15 cm continuous steps
• Motion Host Processor: RK3588 ARM-architecture (1 kHz control frequency)
• Perception Processor: NVIDIA Jetson Xavier NX (8-core ARM A72, 8GB LPDDR4)
• Depth Camera: Intel RealSense D435i RGB-D sensor
• Joint Torque: 50% higher than previous generation
• Visual SLAM Capability: Real-time Visual SLAM using RealSense depth stream
• Neural Network Inference: CUDA-enabled TensorFlow and PyTorch support
• Object Detection: YOLOv8 and other DNN models deployable on Jetson NX
• Operating System: Ubuntu 20.04 (ROS1 and ROS2 compatible)
• ROS Versions: ROS1 (Noetic) and ROS2 (Foxy) with switching support
• Development Access: Full open SDK and API access
• Architecture Design: Split-brain (motion/perception) for research independence
• Positioning: Mid-high tier variant with balanced perception and locomotion
• GPU Architecture: NVIDIA Ampere with 1024 CUDA cores for parallel AI workloads
• Depth Camera Count: Dual Intel RealSense units providing stereo depth perception
• Charging Time: Approximately 1.5 hours from depleted to full charge
• Body Material: High-strength composite chassis with impact-resistant panels and dual camera protection

• Dual-Processor Split-Brain: RK3588 handles motion control; Jetson Xavier NX handles AI perception independently
• NVIDIA Jetson Xavier NX: 8-core ARM A72 processor with dedicated GPU for neural network inference
• Intel RealSense D435i: RGB-D depth camera for visual perception and localization
• Visual SLAM Support: Real-time ORB-SLAM or similar algorithms on depth stream
• Deep Learning Ready: CUDA-enabled TensorFlow and PyTorch for on-robot neural networks
• Object Detection: Run YOLOv8, MobileNet, and other DNN models for visual understanding
• 1 kHz Motion Control: Real-time joint control independent of perception processing
• Reinforcement Learning Support: Train and deploy RL policies via Lite3_rl_deploy
• ROS1 and ROS2 Dual Support: Flexible software framework compatibility with built-in switching
• Enhanced Joint Torque: 50% increased drive torque for improved agility and payload
• Research Independence: Motion stability isolated from perception experiments
• Extended Connectivity: Ethernet connectivity for streaming and network communication
• AI-Optimized Gaits: Pre-loaded RL-trained locomotion algorithms for improved stability
• Modular Sensor Architecture: Support for additional sensors and perception modules
• Enterprise Research Grade: Industrial-strength specifications for academic and commercial research
• Sim-to-Real Pipeline: Support for training in simulation and deploying to hardware
• API for Custom Applications: Full SDK access for developing perception-driven behaviors
• Multi-Framework Compatibility: Supports multiple robotics and AI development frameworks
• Professional Support: Access to dedicated technical resources and research community
• Extensible Design: Open architecture supports future capability additions and upgrades
• 1024 CUDA Cores: Massive parallel processing power for real-time neural network inference and computer vision
• Stereo Depth Perception: Dual RealSense cameras provide overlapping fields of view for robust 3D environment mapping
• Hot-Swappable Battery: Quick-release battery mechanism enables field replacement without tools
• Simultaneous AI Pipelines: GPU headroom supports running perception, planning, and control networks concurrently

The Lite3 Pro is engineered for professional robotics research groups and engineering teams that require a high-performance quadruped platform with the computational headroom for advanced AI workloads. University robotics labs deploy the Lite3 Pro for research in reinforcement learning, sim-to-real transfer, and adaptive locomotion across varied terrain types. The NVIDIA Jetson Orin Nano's GPU acceleration enables real-time neural network inference for perception, planning, and control tasks that would overwhelm CPU-only platforms.

Commercial applications span autonomous inspection of industrial facilities, security patrol in complex indoor-outdoor environments, and development of custom robotic solutions requiring both mobility and onboard intelligence. The dual RealSense depth cameras provide stereo perception for sophisticated obstacle avoidance and environment mapping, while the ROS2 ecosystem compatibility enables rapid integration with existing robotics software stacks. The Lite3 Pro bridges the gap between educational platforms and full industrial-grade quadrupeds, offering professional-tier capability at a research-accessible price point.

The combination of dual depth cameras, GPU computing, and the Lite3 mechanical platform creates a self-contained development system that eliminates external computing dependencies during field testing. Research groups publishing in top-tier robotics venues use the Lite3 Pro to validate perception-locomotion integration approaches that would be prohibitively expensive to develop on larger industrial platforms.

• Estimated Setup Time: 1 to 2 hours for unboxing, charging, component verification, GPU initialization, and running the first autonomous navigation demo.

• Step 1 - Unbox and Verify Components: Confirm all items: Lite3 Pro unit with factory-mounted dual Intel RealSense D435i depth cameras, NVIDIA Jetson Xavier NX perception computer with pre-installed dev environment, rechargeable lithium battery (18 Ah high-capacity), AC charger, Ethernet cable, quick start guide, GPU computing reference, and SDK documentation.

• Step 2 - Charge the Battery: Fully charge the 18 Ah battery (approximately 1.5 hours). This higher-capacity battery provides 2 to 2.5 hours of runtime with cameras and SLAM active, or 4 to 5 hours on standby.

• Step 3 - Power On and Boot: Place the robot on a flat, open surface. Power on and allow approximately 2 minutes for the full boot sequence, including GPU initialization, camera calibration, and sensor validation across the split-brain architecture (RK3588 motion host + Jetson Xavier NX perception host).

• Step 4 - Network Connection: Connect to the Jetson via WiFi or Ethernet (USB 3.0 and HDMI also available). The Pro supports external power input at 24V, 12V, and 5V for extended development sessions.

• Step 5 - Install SDK and ROS Packages: Install the Deep Robotics SDK and ROS2 packages on your development machine. The system runs Ubuntu 20.04 with ROS1 (Noetic) and ROS2 (Foxy) switching support, CUDA-enabled TensorFlow and PyTorch, and YOLOv8.

• Step 6 - Run Diagnostics: Execute the built-in diagnostic suite to verify all 12 DOF, both depth cameras, GPU compute, and sensor fusion are operational.

• Step 7 - First Autonomous Demo: Run the sample autonomous navigation demo to see the stereo depth perception and SLAM capabilities in action. The Pro handles 40-degree slopes, 15 cm stairs, and reaches 4 m/s at full speed.

• IP Rating: IP54 splash-resistant, suitable for outdoor field testing.