Neural Network Robotic Arm Trajectories Under Collision Constraints

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Traditional trajectory optimization methods for robotic arms are suboptimal, leading to slow, jerky movements, unstable grip or suction, and increased chances of errors, as they rely on handcrafted waypoints and require extensive mathematical computations, limiting their ability to adapt to variable scenarios.

Innovation Solution

Utilizing deep neural networks (DNNs) to predict optimized robotic arm trajectories that minimize time, maintain stability, and adhere to collision and kinematic constraints, trained using a variety of scenarios and constraints, including collision-free paths and smooth movements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If traditional trajectory optimization methods using handcrafted waypoints are used, then the system is easier to implement, but the movement speed and smoothness deteriorate

Engineering Contradiction:
Improveease of implementationVSAvoidmovement speed
Core Design Contradiction:
Ease of manufactureVSSpeed

Solution Approach 1:

The patent replaces traditional mathematical computation-based trajectory optimization with a neural network-based system. The neural network is trained offline to learn optimal trajectories, and during runtime, it directly predicts smooth, fast trajectories without requiring extensive real-time mathematical computations, thus substituting the computational mechanics with a learned model.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The neural network is trained in advance using a large dataset of optimized trajectories generated by traditional methods. This preliminary training phase allows the network to internalize optimal movement patterns, so that during actual operation, the network can quickly retrieve and apply pre-learned solutions without performing complex real-time optimization.

Inventive Principle:
Principle #10Preliminary action

2Device complexity

If traditional trajectory optimization methods are used, then the computational approach is simpler, but the adaptability to variable scenarios deteriorates

Engineering Contradiction:
Improvecomputational complexityVSAvoidadaptability to scenarios
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent implements a dynamic trajectory optimization system where the neural network can adapt to different scenarios by taking various inputs (start position, end position, obstacles, task type) and generating appropriate trajectories. The system transitions from static, pre-programmed paths to dynamic, context-aware trajectory generation that can handle variable scenarios.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The neural network is trained on diverse datasets covering multiple scenarios with varying parameters (different obstacles, start/end positions, task types). By learning from this varied training data, the network acquires the ability to generalize and adapt to new scenarios by adjusting its predictions based on the specific parameters of the current situation.

Inventive Principle:
Principle #35Parameter changes

3Quantity of substance

If traditional trajectory optimization methods are used, then the system requires less training data, but the movement smoothness and stability deteriorate

Engineering Contradiction:
Improvetraining data quantityVSAvoidgrip stability
Core Design Contradiction:
Quantity of substanceVSStability of the object's composition

Solution Approach 1:

The training process incorporates feedback mechanisms where trajectories are evaluated based on multiple criteria including smoothness, collision avoidance, and task completion. The loss function provides feedback during training to guide the neural network toward generating trajectories that maintain stability and smoothness, continuously refining the predictions based on performance metrics.

Inventive Principle:
Principle #23Feedback

4Loss of time

If traditional trajectory optimization methods are used, then the computational time per trajectory is longer, but the reliability of trajectory optimization deteriorates

Engineering Contradiction:
Improvecomputational timeVSAvoidtrajectory optimization reliability
Core Design Contradiction:
Loss of timeVSReliability

Solution Approach 1:

The patent replaces complex real-time mathematical optimization with a neural network predictor that has been pre-trained to provide reliable predictions. The network substitutes the traditional optimization engine, delivering both speed and reliability through learned patterns rather than computational search.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The neural network provides continuous, smooth trajectory predictions without the discontinuities and recalculations inherent in traditional methods. Once trained, the network can continuously generate reliable trajectories at high speed, maintaining consistent performance without the intermittent computational bursts required by traditional optimization approaches.

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP4115347B1Trajectory optimization using neural networks
Publication Date: 2025.11.26 EMBODIED INTELLIGENCE INC
  • EP4115347B1 patent drawingFigure 1
  • EP4115347B1 patent drawingFigure 2
  • EP4115347B1 patent drawingFigure 3

AI summary

Various embodiments of the technology described herein generally relate to systems and methods for trajectory optimization with machine learning techniques. More specifically, certain embodiments relate to using neural networks to quickly predict optimized robotic arm trajectories in a variety of scenarios. Systems and methods described herein use deep neural networks to quickly predict optimized robotic arm trajectories according to certain constraints. Optimization, in accordance with some embodiments of the present technology, may include optimizing trajectory geometry and dynamics while satisfying a number of constraints, including staying collision-free, and minimizing the time it takes to complete the task.