Humanoid robotics
Develop human-like locomotion for robots that work around people. Our starting point is walking, running and turning in simulation.
RSO Pose
Trained on human motion capture to predict a body’s next pose and generate natural movement.
- Focus
- Locomotion
- Output
- Pose prediction
- Environment
- Simulation
Learning how a body moves.
We trained our own deep phase motion controller on human motion capture. RSO Pose learns the timing and coordination of movement, then predicts each new pose to make a simulated character walk, run and turn.
Movement in simulation
We examine the movement of the whole body as the controller runs. Joint measurements help us study how the limbs coordinate through each step.
Each pose leads to the next
The controller uses recent movement and the requested direction to predict the next pose. Each new pose becomes part of the history used to generate what follows.
Reading the ground ahead
A terrain scanner could help a robot anticipate changes in the ground before it takes the next step. We are exploring how that information could guide pose prediction.
Movement people can understand
The way a robot moves affects how comfortable people feel around it. We want robots to move in ways people recognise and can anticipate. Learning from human movement gives us a starting point.
Applications
We are developing RSO Pose for uses across robotics and digital characters, starting in simulation.
Robot simulation
Test predicted movement in repeatable virtual environments. Examine transitions, ground contact and the response to a change of direction.
Games
Generate responsive character movement from player input. Explore continuous transitions between walking, running and turning.
Digital avatars
Give virtual characters a natural way of moving. Study how posture, pace and transitions make an avatar feel more lifelike.
Animation tools
Help animators explore movement between poses. Use learned motion as a starting point for sequences they can direct and refine.
Virtual production
Prototype a character’s movement through a scene. Try different paths and pacing while planning a shot or a sequence.
Human–robot interaction
Explore movement people can read and anticipate. Study how a robot approaches, changes direction or comes to a stop around someone.
Testing movement
We test the controller in a live simulation, checking turns, changes of pace and contact with the ground. We refine the generated poses with control and contact adjustments. Applying this work to physical robots will require further development and testing.