Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
 
 
 
 

Repository files navigation

AMD Robotics Hackathon 2025: Selective Waste Sorting via Few Shot Action Learning (Fixed Setup)

Team Information

Team: Team4, Préhistorique

Summary
This repository contains our work for the AMD Robotics Hackathon 2025.

Mission 1 covered hardware and software familiarization (Block Pick and Place).
Mission 2 is our creative project: Selective Waste Sorting (Glass, Plastic, Paper) in a fixed, structured environment, using a Few Shot Action Learning approach built on the LeRobot ACT policy with large external visual priors.


Demo Videos

Mission 1: Block Pick and Place

Mission 2: Waste Sorting Attempts + Failure Analysis

Add your mission 2 video here, ideally showing:

Mission 1

We used the official unified task to complete block pick and place. This helped us quickly familiarize ourselves with:


Mission 2: Selective Waste Sorting

Delivery URLs :

Mission Description

Real world motivation: selective sorting is expensive because collecting large robot-specific datasets is costly and time-consuming.

Our project validates a data-efficient blueprint:

  • learn high-level recognition from public visual datasets
  • learn manipulation from a small local action dataset

Target categories:

  • Glass: bottles, jars
  • Plastic: bottles, packaging, films, trays, bags
  • Paper: cardboard, magazines, office paper

In our fixed setup, the robot attempts this pipeline:

  1. Identify target item class in a cluttered stationary pile
  2. Execute pre-grasp displacement (micro-adjustment for pose refinement)
  3. Grasp
  4. Place into assigned fixed-location bin

This setup isolates failures to the visual-to-action mapping rather than environment randomness.


Creativity / Novelty

1) Few Shot Action Learning with Large Visual Priors

We combine:

  • External visual data (ex: Garbage_Classification_YOLO and similar datasets) to build strong recognition features
  • Minimal robot demonstrations (about 300 action episodes) to learn grasp and place actions

Method:

  1. Train or adapt a visual encoder using public garbage datasets
  2. Freeze this encoder
  3. Fine-tune the ACT action head on our small robot action dataset

Goal: verify whether strong recognition priors reduce the amount of robot action data needed.

2) Pre-Grasp Displacement for Robustness

Instead of directly grasping, the policy outputs a pre-grasp displacement step. In a fixed environment, this improves robustness against small localization errors right before grasp execution.

3) Structured Environment Focus + Failure Mode Analysis

We intentionally constrain:

  • fixed camera placement
  • fixed bin locations
  • fixed workspace layout

This makes errors easier to attribute to:

  • perception mislocalization
  • action head generalization limits
  • contact dynamics / gripper alignment issues

Technical Implementation

Training

  • Policy: ACT (Action Chunking with Transformers) & SmolVLA via LeRobot
  • Inputs: single-view RGB stream + robot joint values
  • Approach: frozen visual encoder + fine-tune action decoder on custom demos

Teleoperation / Dataset Capture

  • Action dataset: about 300 high-quality grasp-and-place demonstrations
  • Purpose: teach the manipulation mapping specific to our robot + environment

Inference and Evaluation

Evaluation focuses on:

  • grasp success rate
  • number of retries needed
  • failure categories (localization vs contact vs kinematics)

Note: the robot did not reach 100% full-task success, which highlights remaining challenges in few-shot action mapping.


How To Reproduce Our Work

Prerequisites

  • AMD environment prepared per Hackathon instructions (ROCm / GPU setup as applicable)
  • conda environment with lerobot installed
  • SO-101 follower and leader arms connected via USB
  • at least one camera connected for recording and inference

Recommended tools:

sudo apt-get update
sudo apt-get install -y v4l-utils

About

AMD robotics hackathon: a weekend spent making a machine move.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors