MOSS: a small, mostly 3D-printed litter-picking rover. Cleaner streets, one maker at a time.
Python
README: OpenAI among friends, simulation and autonomy as work in progress; gripper named for what it is
metroxfcbf91dc2cec
diff --git a/3DPrinting.md b/3DPrinting.mdindex c3d1947..d7ce7ea 100644--- a/3DPrinting.md+++ b/3DPrinting.md@@ -5,7 +5,7 @@| Material | Where | Why ||---|---|---|| PETG | hull, hatch, drivetrain, tensioners, brackets | stiffness and heat near the motors |-| PLA (matte) | cover, bin, arm plate, SO-101 arm, NC90 gripper | big flat parts that print clean |+| PLA (matte) | cover, bin, arm plate, SO-101 arm, gripper | big flat parts that print clean || TPU 95A | tracks, bin pads, gripper pads | grip |Tracks: print with a 0.4 mm nozzle. The 0.8 mm nozzle makes them too stiff to run.@@ -53,9 +53,9 @@ The hull is one piece, 328 mm long: it needs a large bed. A version split for 25| Bin, with magnet pockets | 1 || Arm mounting plate | 1 || SO-101: base motor holder, base, shoulder bracket, shoulder rotation, upper arm, forearm, wrist bracket, wrist | 8 |-| NC90 gripper: base with extended rails, two jaws with extended racks, pinion, two flanges, cable guide, camera mount with 34 mm adapter | 8 |+| Gripper (NormaCore parallel jaw, widened to 90 mm): base with extended rails, two jaws with extended racks, pinion, two flanges, cable guide, camera mount with 34 mm adapter | 8 |-The SO-101 parts are the upstream files with cable clips added. The NC90 jaws and rails are lengthened for bottles and cans.+The SO-101 parts are the upstream files with cable clips added. The gripper is NormaCore's parallel-jaw design with the jaws and rails lengthened to open 90 mm, for bottles and cans.## Tracks and pads, TPU 95Adiff --git a/Assembly.md b/Assembly.mdindex 131845a..1a8dee8 100644--- a/Assembly.md+++ b/Assembly.md@@ -48,7 +48,7 @@ Magnets in the cover and in the bin, pads under the bin.-Build the SO-101 arm with the upstream guide, then the NC90 gripper. Bolt the arm plate to the hull, route the servo bus and the camera cable through the clips.+Build the SO-101 arm with the upstream guide, then the gripper. Bolt the arm plate to the hull, route the servo bus and the camera cable through the clips.## 8. Computer and sensorsdiff --git a/README.md b/README.mdindex e317543..7f9d971 100644--- a/README.md+++ b/README.md@@ -31,7 +31,7 @@ The Jetson configuration is the one on the bench. The Pi configuration is drawn,- Tracked base, two geared DC motors with encoders, printed TPU tracks, an ESP32-S3 driving the motors.- A bin you empty by hand at the end of a run.-- An SO-101 arm (open source) with a NormaCore NC90 gripper and a small camera in the hand.+- An SO-101 arm (open source) with a NormaCore parallel-jaw gripper, widened to a 90 mm grip, and a small camera in the hand.- A depth camera at the nose, a 2D lidar for navigation.- Software: dimOS for teleop and navigation, LeRobot for data and policies. A Raspberry Pi path without dimOS for the first drives.@@ -46,14 +46,16 @@ The rule we build by: pick up more litter with less technology. Every part has t## Friends-[Dimensional](https://x.com/dimensionalos) (dimOS), [NormaCore](https://x.com/norma_core_dev) (gripper), [Tnkr](https://x.com/tnkrdotai) (kits and guide), [Dirac Robotics](https://x.com/diracrobotics) (calibrated simulation), [microduck-lab](https://github.com/jonathanhawkins/microduck-lab) (reinforcement learning). We help each other.+OpenAI (Codex Physical Builds), [Dimensional](https://x.com/dimensionalos) (dimOS), [NormaCore](https://x.com/norma_core_dev) (gripper), [Tnkr](https://x.com/tnkrdotai) (kits and guide), [Dirac Robotics](https://x.com/diracrobotics) (calibrated simulation), [microduck-lab](https://github.com/jonathanhawkins/microduck-lab) (reinforcement learning). We help each other.++## Simulation and autonomy, work in progress++The MOSS body runs in MuJoCo with three recorded pickup missions, in the [moss-jev](https://github.com/metrox-eth/moss-jev) repository. It is a basic model. Dirac Robotics is helping build a calibrated URDF, so the robot in MuJoCo behaves like the real one, to experiment with Dimensional's Dimcode; Dimensional is working on a navigation arena; microduck-lab is trying reinforcement learning on it. Accurate simulation, sim-to-real transfer, a modular stack, autonomy: all of it is part of the journey, none of it is here on day one.## Come and buildAsk anything, tell us what we are doing wrong, it helps the next builder: [Discord](https://discord.com/invite/x2MeNmgveT). The build, as it happens: [@metrox_eth](https://x.com/metrox_eth). The story so far, in a few moments: [STORY.md](STORY.md). What we learned the hard way: [docs/lessons.md](docs/lessons.md). How data from many rovers gets pooled: [docs/data.md](docs/data.md).-Try the simulator: the MOSS body in MuJoCo, with three recorded pickup missions, in the [moss-jev](https://github.com/metrox-eth/moss-jev) repository.-## SafetyThis is a prototype with a lithium battery, motors strong enough to pinch, and no certified protection. Put it on blocks for the first run. Build at your own risk.diff --git a/hardware/README.md b/hardware/README.mdindex 970cc1f..fb9dc55 100644--- a/hardware/README.md+++ b/hardware/README.md@@ -1,6 +1,6 @@# Hardware-Designed in Fusion 360. The STL (one file per part) and STEP files land here at the design freeze, with the fastener list and the sources they derive from (SO-101 arm, NormaCore NC90 gripper) with exact revisions and local changes. The freeze happens after one test on the real robot: a full battery emptied over 3 km on the beach, tracks still on, no bolt lost.+Designed in Fusion 360. The STL (one file per part) and STEP files land here at the design freeze, with the fastener list and the sources they derive from (SO-101 arm, NormaCore parallel-jaw gripper, widened to 90 mm) with exact revisions and local changes. The freeze happens after one test on the real robot: a full battery emptied over 3 km on the beach, tracks still on, no bolt lost.Until then: the [parts list](../3DPrinting.md), the [BOM](../BOM.md) and the photos in `media/` are the reference.