Praxis-1 DROID - Reactor

Praxis-1 DROID predicts a chunk of end-effector pose targets and gripper targets. Its inputs are an exterior camera, a wrist camera, the measured end-effector pose and gripper, and a task instruction. Reactor runs inference, and your application executes the returned targets.

Model information

Property	Value
Model identifier	`reactor/praxis-1-droid`
Family	Praxis-1
Reference embodiment	DROID: seven-joint Franka arm and gripper
Inputs	Two RGB camera tracks, the measured end-effector pose and gripper, and a task instruction
Outputs	Absolute end-effector pose and gripper targets, one row per control step at 15 Hz. A chunk has 24 rows (1.6 seconds of motion).
Session price	$0.030/min, metered per second. See Pricing & Billing.

The model identifier, reactor/praxis-1-droid, selects the DROID variant. Other Praxis-1 checkpoints and robot embodiments do not inherit this observation and action contract. Matching the number of input and output values does not establish compatibility: camera placement, pose frame, gripper convention, and training distribution must match.

Use the model

Start with Get your first actions, then choose simulation or robot integration. The reference specifies the wire contract. Runway’s Praxis-1 repository describes the model and its training.

⌘I