Self-supervised cost of transport estimation for multimodal path planning

Vincent Gherold,Ioannis Mandralis,Eric Sihite,Adarsh Salagame,Alireza Ramezani,Morteza Gharib

2024-12-09

Abstract:Autonomous robots operating in real environments are often faced with decisions on how best to navigate their surroundings. In this work, we address a particular instance of this problem: how can a robot autonomously decide on the energetically optimal path to follow given a high-level objective and information about the surroundings? To tackle this problem we developed a self-supervised learning method that allows the robot to estimate the cost of transport of its surroundings using only vision inputs. We apply our method to the multi-modal mobility morphobot (M4), a robot that can drive, fly, segway, and crawl through its environment. By deploying our system in the real world, we show that our method accurately assigns different cost of transports to various types of environments e.g. grass vs smooth road. We also highlight the low computational cost of our method, which is deployed on an Nvidia Jetson Orin Nano robotic compute unit. We believe that this work will allow multi-modal robotic platforms to unlock their full potential for navigation and exploration tasks.

Robotics,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to enable a robot to autonomously decide to select the path with the optimal energy consumption for navigation given high - level goals and environmental information. Specifically, the author proposes a self - supervised learning method, which enables the robot to estimate the Cost of Transport (COT) of its surrounding environment only through visual input, thereby helping multi - modal robot platforms unlock their full potential in navigation and exploration tasks.

### Main Problem Description

1. **Selection of the Path with Optimal Energy Consumption**:
- The robot needs to autonomously select the path with the minimum energy consumption in a complex environment.
- This involves the energy consumption assessment of different terrains (such as grasslands, smooth roads, etc.).

2. **Application of Multi - Modal Robots**:
- Multi - modal robots (such as the M4 robot) can move in different modes (such as driving, flying, walking, etc.).
- The energy consumption differences in different modes are significant, so an effective method is required to assess and select the optimal path.

### Solution Overview

To achieve this goal, the author has developed a method based on self - supervised learning. It uses RGB - D images as input to predict the COT value of each pixel and projects these values into a Bird - Eye - View (BEV). This method allows the robot to assess the energy consumption of different terrains in real - time and plan the optimal path according to these assessment results.

### Formula Explanation

COT (Cost of Transport) is defined as follows:

\[ \text{COT}=\frac{E}{mgd}=\frac{P}{mgv} \]

where:
- \( E \) is the total energy consumed by the robot,
- \( m \) is the mass of the robot,
- \( g \) is the acceleration due to gravity,
- \( d \) is the distance traveled,
- \( P \) is the power consumption,
- \( v \) is the speed of the robot.

This formula is used to quantify the energy efficiency of the robot moving on different terrains and is a dimensionless and continuous indicator.

### Method Advantages

1. **Real - Time Deployment**: By predicting 2D COT images, the task of the model is simplified, making it easier to be deployed in real - time.
2. **Efficiency**: This method has a low computational cost and is suitable for embedded computing units (such as Nvidia Jetson Orin Nano).
3. **Universality**: It can be applied to any robot platform with an RGB - D sensor.

In conclusion, this paper aims to enable robots to autonomously assess the energy consumption of different terrains and select the optimal path through the self - supervised learning method, thereby improving the navigation and exploration capabilities of multi - modal robots in complex environments.