NRN’s approach to robotic sports is grounded in progressive, real-world experimentation. Our roadmap is structured into phases, each designed to validate and evolve the capabilities of our Sim-to-Real reinforcement learning (RL) pipeline. From robotic arms to humanoids and racing drones, we aim to demonstrate the full range of embodied AI in competitive settings.
In this initial phase, we focus on showcasing the integrity of our RL-powered continuous learning pipeline using a robotic arm named RME-1 (pronounced “Arm-y 1”). This stage is centered around proving core capabilities like data collection and real-time learning. Demonstrations will include:
Object pickup and manipulation tasks
Stacking and fine motor control
Mini-putt challenges to showcase an understanding of environmental physics
Dynamic sparring drills to highlight reaction time and motion prediction
These tests serve as the foundation for more complex embodied AI behavior and validate our client-side data collection tools and training infrastructure in real-world conditions.
Building on the success of RME-1, we will apply the full NRN RL pipeline to humanoid agents.
Begin with miniature humanoid robots to test RL agent performance in bipedal combat
Launch a full robotic combat competition campaign featuring humanoid tournaments
Scale toward larger, more complex humanoids with expanded mobility and dexterity
Long-term milestone: Develop and deploy full-sized humanoid competitors trained via continual learning, capable of dynamic physical interaction in competitive matches
In addition to robotic combat, we will expand the NRN platform to other categories of robotic competition, each emphasizing different skill domains and control systems:
Robot Dog Racing & Challenges: Quadrupedal agents showcasing agility and terrain adaptation
Robot Kart Racing: Fast-paced, agent-controlled vehicles navigating real tracks
Drone Racing: Aerial agents trained on trajectory prediction and reactive flight control
Robot Athletics: Obstacle courses, climbing, jumps—pushing physical versatility and sim-to-real transfer