Multi-Object Robot Navigation Using Semantic Maps and DRL

Solution Overview

Problem

Existing methods for object navigation in unseen environments struggle to generalize and utilize prior knowledge or semantic context, leading to difficulties in tracking and navigating multiple target objects without pre-sequencing.

Innovation Solution

A system and method utilizing a deep reinforcement learning framework that employs a semantic map to identify target objects, computes high-level feature embeddings, and uses an actor-critic network to navigate a shortest path to multiple objects, rewarding concurrent progress and leveraging prior knowledge.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If Reinforcement Learning techniques are employed for mapping pixels directly to actions, then navigation actions can be generated, but the system finds it difficult to generalize to previously unseen scenarios since it lacks to build a representation of the unseen environment

Engineering Contradiction:

Improvegeneralization capabilityVSAvoidenvironment representation

Core Design Contradiction:

Adaptability or versatilityVSLoss of information

Solution Approach 1:

The system performs preliminary actions by building a semantic map of the environment before navigation tasks are executed. This semantic map serves as prior knowledge that is constructed in advance, allowing the system to generalize to unseen scenarios without requiring complete relearning. The semantic map includes object locations, semantic categories, and spatial relationships that are prepared beforehand.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The semantic map acts as an intermediary between raw sensory inputs and navigation decisions. Instead of mapping pixels directly to actions, the system first creates a semantic representation that mediates the navigation process. This intermediary representation captures essential environmental features while filtering out irrelevant details, enabling better generalization.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If the assistive agent implicitly learns the navigation problem through recursive trials, then navigation capability is developed, but the process requires extensive trials and lacks efficiency in locating target objects

Engineering Contradiction:

Improvenavigation capabilityVSAvoidlearning time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by building a semantic map of the environment before navigation tasks are executed. This semantic map serves as prior knowledge that is constructed in advance, allowing the system to generalize to unseen scenarios without requiring complete relearning. The semantic map includes object locations, semantic categories, and spatial relationships that are prepared beforehand.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system uses feedback mechanisms where the semantic map is continuously updated based on sensor observations and navigation outcomes. This feedback loop allows the system to refine its environmental representation efficiently, reducing the number of trials needed for successful navigation by learning from past experiences and incorporating that knowledge into the semantic map.

Inventive Principle:

Principle #23Feedback

3Measurement precision

If existing methods locate the target object by finding similarity between current observation and the target observation through a trained Siamese Network, then object location is achieved, but the methods lack in utilizing prior knowledge or semantic context to locate specific target object in the unseen environment

Engineering Contradiction:

Improveobject location accuracyVSAvoidutilization of prior knowledge

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary actions by building a semantic map of the environment before navigation tasks are executed. This semantic map serves as prior knowledge that is constructed in advance, allowing the system to generalize to unseen scenarios without requiring complete relearning. The semantic map includes object locations, semantic categories, and spatial relationships that are prepared beforehand.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system changes the parameter of object representation from raw pixel data to semantic features extracted by the encoder network. This transformation allows the system to utilize semantic context and prior knowledge stored in the semantic map, improving adaptability to unseen environments while maintaining location accuracy through meaningful feature comparisons.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12597267B2Method and system for multi-object tracking and navigation without pre-sequencing

Publication Date: 2026.04.07 TATA CONSULTANCY SERVICES LTD

AI summary

This disclosure relates generally to method and system for multi-object tracking and navigation without pre-sequencing. Multi-object navigation is an embodied AI task where object navigation only searches for an instance of at least one target object where a robot localizes an instance to locate target objects associated with an environment. The method of the present disclosure employs a deep reinforcement learning (DRL) based framework for sequence agnostic multi-object navigation. The robot receives from an actor critic network a deterministic local policy to compute a low-level navigational action to navigate along a shortest path calculated from a current location of the robot to the long-term goal to reach the target object. Here, a deep reinforcement learning network is trained to assign the robot with a computed reward function when the navigational action is performed by the robot to reach an instance of the plurality of target objects.