RAMP: Hierarchical Reactive Motion Planning for Manipulation Tasks Using Implicit Signed Distance Functions

We introduce Reactive Action and Motion Planner (RAMP), a hierarchical approach where a novel variant of a Model Predictive Path Integral (MPPI) controller is used to generate trajectories which are then followed asynchronously by a local vector field controller. We demonstrate that RAMP can rapidly find paths in the robot's configuration space, satisfy task and robot-specific constraints, and provide safety by reacting to static or dynamically moving obstacles.

RAMP achieves superior performance through a number of key innovations: we use Signed Distance Function (SDF) representations directly from the robot configuration space, both for collision checking and reactive control. The use of SDFs allows for a smoother definition of collision cost when planning for a trajectory, and is critical in ensuring safety while following trajectories. In addition, we introduce a novel variant of MPPI which, combined with the safety guarantees of the vector field trajectory follower, performs incremental real-time global trajectory planning.

Simulation results establish that our method can generate paths that are comparable to traditional and state-of-the-art approaches in terms of total trajectory length while being up to 30 times faster. Real-world experiments demonstrate the safety and effectiveness of our approach in challenging table clearing scenarios.

Our motion planning architecture consists of two distinct modules running in parallel and communicating asynchronously: an MPPI-based trajectory generator and a vector field-based trajectory follower. Both modules use estimates of the robot body's signed distance to the scene point cloud given its joint configuration. We refer to these values as the Configuration Signed Distance Function (C-SDF) values. The trajectory generator uses C-SDF values to estimate the collision cost of proposed states during planning, whereas the trajectory follower uses the C-SDF value and gradient with respect to the current configuration in order to avoid obstacles. With GPU acceleration, we can afford to use an accurate C-SDF estimation algorithm, based on direct computation of distances between the robot and the scene.

The robot successfully completes grasping, leftover disposal and dishwasher placing tasks while (in real-time) avoiding static and dynamic objects in the scene.

The robot is commanded to navigate to specific targets (each visualized by a red marker) from its current position (green marker), while avoiding objects on the table and a moving ball (in blue).