Jellyfishbot

The Jellyfishbot is a commercial autonomous surface vessel built by IADYS, a French company specialized in robotic solutions for water depollution. Unlike a research platform, the Jellyfishbot is designed for real-world operations in ports, marinas, and industrial sites, collecting floating waste and hydrocarbons either teleoperated or in autonomous mode.

We use it in the context of the R3AMA ANR project, which focuses on robust reinforcement learning for floating waste collection.

A more capable platform

Compared to the Kingfisher I used during my PhD, the Jellyfishbot is a more mature, production-grade platform. Three thrusters for omnidirectional motion in tight spaces, longer battery life, and a sensor and software stack already designed around real cleaning operations. It’s built to be deployed and used by non-roboticists, which is a very different design approach from a research ASV.

Adding a research layer on top

The trade-off with a commercial platform is that the internals are proprietary, so we don’t get direct access to the robot’s embedded computer. To work around this, we added a separate waterproof enclosure on top of the Jellyfishbot, hosting our own compute (NVIDIA Jetson Orin Nano). This gives us a ROS-based interface for prototyping new perception and control algorithms, while leaving the manufacturer’s stack untouched.

What’s next

The next step is to port the RL-based control pipeline previously developed for the Kingfisher onto the Jellyfishbot, and extend it with available sensing modalities, in particular underwater sonar (for bathymetry and submerged obstacle awareness) and lidar-based obstacle avoidance for navigating safely around docks, boats, and other structures.