ScheduleStream: Temporal Planning with Samplers for GPU-Accelerated Multi-Arm Task and Motion Planning & Scheduling
Caelan Garrett, Fabio Ramos
AI summary
Problem
Most existing TAMP algorithms produce serial plans where only one arm moves at a time, preventing efficient parallel multi-arm manipulation. There is a lack of general-purpose frameworks that integrate temporal scheduling with continuous sampling and GPU acceleration.
Approach
The authors introduce ScheduleStream, a Python-based framework that models robot actions as hybrid durative events and uses procedural samplers to generate continuous parameters. Its algorithms alternate between scheduling and sampling phases to lazily construct and optimize parallel action schedules, leveraging GPU acceleration for kinematics and collision checking.
Key results
- First domain-independent temporal planning language with procedural samplers
- Novel lazy scheduling algorithms that minimize schedule makespan
- GPU-accelerated sampling integration for Task and Motion Planning & Scheduling
- Higher success rates and lower makespans in simulation, plus real-world bimanual robot demonstrations
Why it matters
Provides a scalable, domain-independent solution for parallel multi-arm control, advancing automation for bimanual and humanoid robots in industrial and home environments.
Abstract
Bimanual and humanoid robots are appealing because of their human-like ability to leverage multiple arms to efficiently complete tasks. However, controlling multiple arms at once is computationally challenging due to the growth in the hybrid discrete-continuous action space. Task and Motion Plan- ning (TAMP) algorithms can efficiently plan in hybrid spaces but generally produce plans, where only one arm is moving at a time, rather than schedules that allow for parallel arm motion. In order to extend TAMP to produce schedules, we present ScheduleStream, the first general-purpose framework for plan- ning & scheduling with sampling operations. ScheduleStream models temporal dynamics using hybrid durative actions, which can be started asynchronously and persist for a duration that’s a function of their parameters. We propose domain-independent algorithms that solve ScheduleStream problems without any application-specific mechanisms. We apply ScheduleStream to Task and Motion Planning & Scheduling (TAMPAS), where we use GPU acceleration within samplers to expedite planning. We compare ScheduleStream algorithms to several ablations in simulation and find that they produce more efficient solutions. We demonstrate ScheduleStream on several real-world biman- ual robot tasks at https://schedulestream.github.io.